Multi-level gridding deployment method, system and equipment based on hierarchical clustering division

Abstract

The invention discloses a hierarchical clustering division-based multi-level gridding deployment method, system and device, and the method comprises the steps: for any inspection point pair, calculating a three-dimensional geographic curved surface flight distance, and calculating a communication signal attenuation prediction value; constructing a multi-dimensional flight cost matrix to obtain an initial inspection grid; determining a power grid inspection point, a risk device and a normal device with a fault based on the device fault alarm information; defining a power grid inspection point with a fault as a potential energy source, defining risk equipment as a low potential well, setting potential energy of normal equipment as 0, and generating a real-time fault potential field distribution diagram; and calculating a total fault entropy value based on the real-time fault potential field distribution diagram, evaluating a predefined grid operation set, selecting an operation which maximizes an entropy deceleration rate as an optimal decision, outputting a new grid boundary scheme, generating a collaborative inspection task for the unmanned aerial vehicle cluster in each new grid, and issuing and executing the collaborative inspection task. The method is used for realizing efficient inspection response to the dynamic fault zone, and the inspection efficiency is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) inspection technology, and in particular to a multi-level gridded deployment method, system, and equipment based on hierarchical clustering. Background Technology

[0002] With the deepening of smart grid construction, drone inspection has become an important means of power system operation and maintenance. Regular inspections of key equipment such as transmission lines and substations using drone swarms can effectively improve inspection efficiency and reduce labor costs. In practical applications, to rationally allocate drone resources, a grid-based deployment method is typically adopted, dividing the inspection area into several grids, with each grid assigned a corresponding drone to perform inspection tasks.

[0003] Existing methods for delineating drone inspection grids primarily rely on static divisions based on geographical distance or administrative boundaries. Under normal operating conditions, these methods can adequately meet daily inspection needs. However, when the power grid faces extreme weather (such as typhoons or freezing disasters) or cascading equipment failures, the faults are often not isolated events but propagate rapidly along the electrical topology and geographical space of the power grid, forming dynamically evolving fault zones. This cascading effect of faults quickly renders even the most reasonable grid delineation schemes ineffective. Specifically, when a fault spreads across multiple grids, drones within each grid continue to operate independently according to their original boundaries. Drone resources in the fault origin area may already be saturated, while areas where the fault is about to spread are still performing routine inspection tasks, resulting in a severe misallocation of emergency resources and missing the golden time for fault response.

[0004] In summary, current methods for deploying drones in a grid-like manner are insufficient for efficiently responding to dynamic fault zones when dealing with scenarios involving cascading fault propagation, resulting in low inspection efficiency. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned defects and problems in the prior art and provide a multi-level gridded deployment method, system and equipment based on hierarchical clustering, which can realize efficient inspection response to dynamic fault zones and effectively improve inspection efficiency.

[0006] To achieve the above objectives, the technical solution of the present invention is:

[0007] In a first aspect, the present invention provides a multi-level gridded deployment method based on hierarchical clustering, applied to a power grid drone inspection system, the power grid drone inspection system including a drone swarm, the method comprising:

[0008] Obtain digital elevation model data of the inspection area in the power grid inspection points;

[0009] For any pair of inspection points, calculate the flight distance on the three-dimensional geographic surface, and calculate the predicted value of communication signal attenuation based on digital elevation model data and a preset wireless channel propagation model.

[0010] A multidimensional flight cost matrix is ​​constructed based on the predicted flight distance and communication signal attenuation values ​​of three-dimensional geographic surface.

[0011] The multidimensional flight cost matrix is ​​used as input to aggregate all inspection points to obtain the initial inspection grid.

[0012] Real-time acquisition of equipment fault alarm information, and based on the equipment fault alarm information, identification of power grid inspection points where faults have occurred, risky equipment, and normal equipment;

[0013] Define the faulty power grid inspection point as a potential energy source and set an initial potential energy value; define the risky equipment as a low potential sink and set a potential energy value; set the potential energy of normal equipment to 0.

[0014] The electrical topology and geographical proximity of power grid inspection points are obtained, and a real-time fault potential field distribution map is generated based on the electrical topology and geographical proximity to simulate the process of potential energy spreading from the potential energy source to the surrounding area.

[0015] Based on the real-time fault potential field distribution map, the total fault entropy value of the current power grid system is calculated, and the predefined set of grid operations is evaluated based on the total fault entropy value. The operation that maximizes the entropy deceleration rate is selected as the optimal decision, and the adjusted new grid boundary scheme is output.

[0016] Based on the new grid boundary scheme, collaborative inspection tasks are generated for each drone cluster within the new grid and then issued for execution.

[0017] Preferably, the construction of a multidimensional flight cost matrix based on the predicted flight distance and communication signal attenuation values ​​of the three-dimensional geographic surface includes:

[0018] For any pair of inspection points, the first weighted value is obtained by multiplying the flight distance of the three-dimensional geographic surface by the preset flight distance weighting coefficient, and the second weighted value is obtained by multiplying the predicted value of communication signal attenuation by the preset communication attenuation weighting coefficient.

[0019] The sum of the first weighted value and the second weighted value yields the comprehensive flight cost of the inspection point pair;

[0020] The comprehensive flight costs of all inspection point pairs are organized into a matrix form to obtain a multidimensional flight cost matrix.

[0021] Preferably, the step of aggregating all inspection points using the multidimensional flight cost matrix as input to obtain the initial inspection grid includes:

[0022] Initialize the clustering state, treating each inspection point as an independent cluster;

[0023] Repeat the aggregation step until the number of clusters reaches the preset grid number threshold;

[0024] Each resulting cluster is defined as the initial inspection grid.

[0025] The polymerization step includes:

[0026] Calculate the distance between any two clusters based on the multidimensional flight cost matrix;

[0027] The two clusters with the smallest distance are selected and merged to form a new cluster;

[0028] Update the multidimensional flight cost matrix, recalculate the inter-cluster distances that include the new clusters, and determine the number of current clusters based on the inter-cluster distances.

[0029] Preferably, the step of determining the power grid inspection point where the fault occurred, the at-risk equipment, and the normal equipment based on equipment fault alarm information includes:

[0030] Analyze the equipment fault alarm information to obtain the location information and fault type of the faulty equipment;

[0031] The fault level is queried from the preset fault level mapping table according to the fault type, and the power grid inspection point where the fault occurred is determined based on the fault level.

[0032] Obtain the operating years and historical fault frequency of each device at the power grid inspection point where a fault occurred, and calculate the risk score of each device;

[0033] Devices with risk scores exceeding a preset risk threshold are identified as risky devices, while devices with risk scores below the preset risk threshold are identified as normal devices.

[0034] Preferably, generating a real-time fault potential field distribution map based on electrical topology and geographical proximity includes:

[0035] Construct a power grid topology graph, where nodes represent power grid inspection points and edges represent electrical topology or geographical proximity relationships;

[0036] Assign a diffusion coefficient to each edge of the power grid topology graph, where the diffusion coefficient corresponding to electrical topology is greater than the diffusion coefficient corresponding to geographical proximity.

[0037] Based on the initial potential energy value and diffusion coefficient, calculate the potential energy value of each node in the power grid topology at each time step;

[0038] The iteration stops when the number of iterations reaches the preset iteration threshold or the potential energy distribution converges, and the final potential energy distribution is obtained.

[0039] The final potential energy distribution is mapped to a preset power grid geographic space to generate a real-time fault potential field distribution map.

[0040] Preferably, the step of calculating the potential energy value of each node in the power grid topology at each time step based on the initial potential energy value and the diffusion coefficient includes:

[0041] For any node, obtain all neighboring nodes that are adjacent to that node;

[0042] Calculate the difference between the potential energy value of the neighboring node at the current time step and the potential energy value of this node at the current time step;

[0043] Multiply the difference in potential energy values ​​by the diffusion coefficient of the corresponding edge to obtain the potential energy transfer amount;

[0044] Sum the potential energy transferred from all neighboring nodes to this node, and add the potential energy value of this node at the current time step to obtain the potential energy value of this node at the next time step.

[0045] Preferably, the power grid drone inspection system further includes a ground base station connected to the drone cluster. The ground base station generates and executes collaborative inspection tasks for each drone cluster within a new grid based on the new grid boundary scheme, including:

[0046] The new grid boundary scheme is analyzed to obtain the boundary range of each new grid and the UAV resources within the grid, and the UAV endurance is determined based on the UAV resources;

[0047] Based on the real-time fault potential field distribution map, high-risk areas with potential energy values ​​exceeding the preset potential energy threshold are identified in each new grid, and the potential energy gradient of the high-risk areas is determined.

[0048] For each new grid, based on the potential energy gradient of high-risk areas and the drone's endurance, a collaborative inspection path is planned for the drone swarm within the new grid;

[0049] The collaborative inspection path is encapsulated as an inspection task instruction and sent to the corresponding drone cluster via ground base stations.

[0050] Preferably, the step of planning a collaborative inspection path for the drone swarm within the new grid based on the potential energy gradient of the high-risk area and the drone's endurance includes:

[0051] Based on the potential energy gradient in high-risk areas, determine the direction of the fastest increase in potential energy.

[0052] Determine the priority inspection order based on the direction of the fastest increase in potential energy.

[0053] Obtain the current location, remaining battery power, and maximum range of all drones within the new grid;

[0054] Select inspection points in order of priority and calculate the flight distance of each drone from its current position to the inspection point.

[0055] Repeat the path planning steps until all inspection points have been assigned or all drones have run out of battery life, generating a complete inspection path for each drone in the drone swarm.

[0056] The path planning steps include:

[0057] For the current inspection point, the flight distance is compared with the maximum range of the drone to select candidate drones whose range meets the requirements;

[0058] Select the drone with the shortest flight distance from the candidate drones, assign the inspection point to that drone, and update the drone's remaining range.

[0059] Secondly, the present invention provides a multi-level gridded deployment system based on hierarchical clustering, the system being used to implement the method described above, the system comprising:

[0060] The initial inspection grid acquisition module is used to acquire digital elevation model data of the inspection area in the power grid inspection points; for any pair of inspection points, it calculates the three-dimensional geographic surface flight distance and calculates the communication signal attenuation prediction value based on the digital elevation model data and the preset wireless channel propagation model; based on the three-dimensional geographic surface flight distance and the communication signal attenuation prediction value, it constructs a multi-dimensional flight cost matrix; and aggregates all inspection points using the multi-dimensional flight cost matrix as input to obtain the initial inspection grid.

[0061] The real-time fault potential field distribution map acquisition module is used to acquire equipment fault alarm information in real time, and based on the equipment fault alarm information, determine the power grid inspection point where the fault occurred, the risk equipment, and the normal equipment; define the power grid inspection point where the fault occurred as a potential energy source and set an initial potential energy value, define the risk equipment as a low potential well and set a potential energy value, and set the potential energy of the normal equipment to 0; acquire the electrical topology and geographical proximity of the power grid inspection point, and generate a real-time fault potential field distribution map based on the electrical topology and geographical proximity to simulate the process of potential energy spreading from the potential energy source to the surrounding area;

[0062] The new inspection grid acquisition module is used to calculate the total fault entropy value of the current power grid system based on the real-time fault potential field distribution map, evaluate the predefined grid operation set based on the total fault entropy value, select the operation that maximizes the entropy deceleration rate as the optimal decision, and output the adjusted new grid boundary scheme; according to the new grid boundary scheme, it generates collaborative inspection tasks for the UAV cluster in each new grid and issues them for execution.

[0063] Thirdly, the present invention provides a multi-level gridded deployment device based on hierarchical clustering, including a memory and a processor;

[0064] The memory is used to store computer program code and transmit the computer program code to the processor;

[0065] The processor is configured to execute the method described above according to instructions in the computer program code.

[0066] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0067] This invention discloses a multi-level gridded deployment method, system, and device based on hierarchical clustering. Through the aforementioned technical solution, it effectively solves the resource mismatch problem of traditional static grid partitioning in scenarios involving cascading fault propagation. By acquiring digital elevation model data and calculating predicted flight distances and communication signal attenuation values ​​on three-dimensional geographic surfaces, the constructed multi-dimensional flight cost matrix fully considers the impact of terrain factors and communication quality on UAV inspections, making the initial grid partitioning more scientific and reasonable, avoiding the limitations of traditional methods that only consider planar distances. Using a hierarchical clustering algorithm for grid partitioning can adaptively aggregate inspection points with similar flight costs and communication quality into the same grid, improving the collaborative efficiency within the grid. More importantly, by acquiring equipment fault alarm information in real time and constructing a fault potential energy model and a real-time fault potential field distribution map, it is possible to intuitively predict the spatial distribution and evolution trend of fault risks. Based on a dynamic grid adjustment mechanism using fault entropy values, by evaluating the impact of different grid operations on the system entropy value, the operation that maximizes the entropy deceleration rate is selected as the optimal decision, realizing intelligent dynamic adjustment of grid boundaries. This allows UAV resources to be flexibly redeployed as faults evolve, effectively avoiding waste and shortages of emergency resources. Based on the adjusted new grid boundary scheme and combined with potential energy gradient information, collaborative inspection tasks are generated for the UAV swarm, ensuring that high-risk areas receive priority and intensive inspection coverage. This significantly improves the timeliness of fault detection and handling, thereby increasing inspection efficiency. It is not only suitable for emergency scenarios such as extreme weather or cascading failures, but also enables dynamic optimization of inspection resource allocation based on equipment status during daily operation, comprehensively enhancing the intelligence level and emergency response capabilities of the power grid UAV inspection system. Attached Figure Description

[0068] Figure 1 This is a flowchart of a multi-level gridded deployment method based on hierarchical clustering provided in an embodiment of the present invention.

[0069] Figure 2 This is a schematic diagram illustrating the calculation process of inter-cluster distance provided in an embodiment of the present invention.

[0070] Figure 3This is a structural block diagram of a multi-level gridded deployment system based on hierarchical clustering provided in an embodiment of the present invention.

[0071] Figure 4 This is a structural block diagram of a multi-level gridded deployment device based on hierarchical clustering provided in an embodiment of the present invention. Detailed Implementation

[0072] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0073] See Figure 1 This invention provides a multi-level gridded deployment method based on hierarchical clustering, applied to a power grid drone inspection system, which includes a drone swarm. The method includes the following steps.

[0074] S110. Obtain digital elevation model data of the inspection area in the power grid inspection point;

[0075] S120. For any pair of inspection points, calculate the flight distance of the three-dimensional geographic surface, and calculate the predicted value of communication signal attenuation based on digital elevation model data and preset wireless channel propagation model.

[0076] S130. Construct a multidimensional flight cost matrix based on the predicted flight distance and communication signal attenuation values ​​of the three-dimensional geographic surface.

[0077] S140. Aggregate all inspection points using the multidimensional flight cost matrix as input to obtain the initial inspection grid.

[0078] S150: Acquire equipment fault alarm information in real time, and determine the power grid inspection point where the fault occurred, the risk equipment, and the normal equipment based on the equipment fault alarm information;

[0079] S160. Define the faulty power grid inspection point as a potential energy source and set an initial potential energy value, define the risky equipment as a low potential sink and set a potential energy value, and set the potential energy of normal equipment to 0.

[0080] S170. Obtain the electrical topology and geographical proximity of the power grid inspection points, and generate a real-time fault potential field distribution map based on the electrical topology and geographical proximity to simulate the process of potential energy spreading from the potential energy source to the surrounding area.

[0081] S180. Based on the real-time fault potential field distribution map, calculate the total fault entropy value of the current power grid system, evaluate the predefined set of grid operations based on the total fault entropy value, select the operation that maximizes the entropy deceleration rate as the optimal decision, and output the adjusted new grid boundary scheme.

[0082] S190. Generate collaborative inspection tasks for each drone cluster within the new grid according to the new grid boundary scheme and issue them for execution.

[0083] In this embodiment, the total fault entropy value of the current power grid system is calculated based on the real-time fault potential field distribution map, including:

[0084] The real-time fault potential field distribution map is divided into multiple spatial units;

[0085] Calculate the potential energy value within each spatial unit and then calculate the total potential energy of all spatial units.

[0086] Calculate the proportion of the potential energy value of each spatial unit to the total potential energy to obtain the potential energy distribution probability;

[0087] Based on the potential energy distribution probability, the total fault entropy value of the current power grid system is calculated using the information entropy calculation formula.

[0088] The predefined set of mesh operations includes operations to merge adjacent meshes, operations to split the current mesh, and operations to exchange parts of the mesh.

[0089] The predefined set of mesh operations is evaluated based on the total fault entropy value. The operation that maximizes the entropy deceleration rate is selected as the optimal decision, and the adjusted new mesh boundary scheme is output, including the following steps:

[0090] For each candidate operation in the set of mesh operations, simulate the mesh state after executing the operation;

[0091] Based on the grid state after the simulation, the fault potential field distribution is recalculated, and the estimated fault entropy value under the simulation state is calculated.

[0092] Calculate the entropy deceleration rate for each candidate operation. The entropy deceleration rate is the difference between the current total fault entropy value and the estimated fault entropy value divided by a preset time interval.

[0093] Choose the candidate operation with the largest entropy deceleration rate as the optimal decision;

[0094] Adjust the grid boundary based on the optimal decision and output the new grid boundary scheme.

[0095] Following S190, the following steps are also included:

[0096] Receive inspection result data fed back by the drone after it has performed the inspection mission. The inspection result data includes newly discovered fault information or area safety confirmation information.

[0097] Update the real-time fault potential field distribution map based on the inspection results data;

[0098] Based on the updated real-time fault potential field distribution map, S180 is re-executed.

[0099] In this embodiment, the drone swarm consists of multiple industrial-grade drones equipped with sensors such as high-definition cameras, infrared thermal imagers, and lidar, enabling comprehensive inspections of power facilities such as transmission lines, substations, and towers. The ground base station communicates with the drones, receives inspection data, and issues task commands. In practice, the first step is to acquire digital elevation model (DEM) data for the power grid inspection area. A DEM is a digital representation of the three-dimensional terrain, recording the altitude information of each geographical location within the inspection area. This acquisition can be achieved through satellite remote sensing data, aerial photogrammetry, lidar scanning, etc. For large-scale power grid inspection areas, publicly available geographic information databases can be used to obtain DEM data. This data is typically stored in raster format, with each raster cell containing the latitude and longitude coordinates and altitude value of that location.

[0100] After acquiring the raw digital elevation model data, preprocessing operations are performed, including coordinate system transformation, data accuracy calibration, and missing value imputation, to ensure consistency with the coordinate system of the UAV navigation system. For missing or outlier values ​​in the data, nearest neighbor interpolation or kriging interpolation can be used to repair them, ensuring data continuity and integrity.

[0101] In addition, data cropping is required based on the actual extent of the inspection area to extract elevation data for the smallest bounding rectangle covering all power grid inspection points, thereby reducing the amount of data required for subsequent calculations. The digital elevation model data obtained through this step provides fundamental terrain information support for subsequent calculations of three-dimensional flight distance and communication signal attenuation, ensuring that the grid division fully considers the impact of terrain factors on UAV flight and communication.

[0102] After acquiring the digital elevation model data, it is necessary to calculate the three-dimensional geographic surface flight distance and the predicted communication signal attenuation between any pair of inspection points. The calculation of the three-dimensional geographic surface flight distance differs from the traditional two-dimensional plane straight-line distance calculation; the impact of terrain undulations on the flight path must be considered.

[0103] In practice, the first step is to extract a surface profile line connecting the two inspection points from the digital elevation model. This profile line consists of a series of discrete elevation sampling points. Then, the three-dimensional spatial distance between adjacent sampling points is calculated, considering both horizontal distance and vertical height difference. By summing the three-dimensional distances between all adjacent sampling points, the total flight distance along the surface is obtained. For example, assuming the profile line between inspection point A and inspection point B contains 100 sampling points, the horizontal distance between the i-th sampling point and the (i+1)-th sampling point is... The vertical height difference is The three-dimensional distance between the two points is: ;

[0104] For calculating the predicted attenuation of communication signals, a pre-defined wireless channel propagation model is used, such as the free-space propagation model or the logarithmic distance path loss model. In practical applications, considering the impact of terrain obstruction on signal propagation, it is necessary to determine whether a line-of-sight communication path exists between two inspection points. By analyzing digital elevation model data, it is checked whether the straight line connecting the two points is obstructed by the terrain in between. If obstruction exists, diffraction loss or reflection loss needs to be calculated in the propagation model. For example, when the signal needs to propagate around a mountain peak, a knife-edge diffraction model can be used to calculate the additional attenuation.

[0105] By comprehensively considering factors such as propagation distance, terrain obstruction, and antenna height, the predicted value of communication signal attenuation between the two inspection points is finally obtained. This value reflects the ease with which the UAV can maintain stable communication with the ground base station while flying between the two points. This calculation provides key distance and communication quality parameters for the subsequent construction of a multidimensional flight cost matrix.

[0106] Based on the calculated flight distance and communication signal attenuation predictions from the three-dimensional geographic surface, a multi-dimensional flight cost matrix is ​​constructed. This matrix comprehensively considers both flight distance cost and communication quality cost, enabling a more comprehensive assessment of the correlation strength between inspection points.

[0107] In practical implementation, for any pair of inspection points, the flight distance on the 3D geographic surface is first multiplied by a preset flight distance weighting coefficient to obtain a first weighted value. This weighting coefficient reflects the importance of flight distance in the total cost. Then, the predicted communication signal attenuation value is multiplied by a preset communication attenuation weighting coefficient to obtain a second weighted value, which reflects the importance of communication quality. The settings of these two weighting coefficients need to be adjusted according to the actual application scenario. For example, in mountainous areas with complex terrain and poor communication conditions, the communication attenuation weighting coefficient should be set higher to ensure that the UAV can maintain stable data transmission.

[0108] The sum of the first and second weighted values ​​yields the comprehensive flight cost of the inspection point pair. A smaller cost value indicates a stronger correlation between the two inspection points, making them more suitable for being assigned to the same grid. By traversing all possible combinations of inspection point pairs, the comprehensive flight cost of each pair is calculated, and these cost values ​​are organized into a matrix to obtain a multidimensional flight cost matrix. This matrix is ​​symmetric, with rows and columns representing different inspection points, and matrix elements representing the comprehensive flight cost of the corresponding inspection point pair. For example, assuming the inspection area contains 50 inspection points, the multidimensional flight cost matrix is ​​a 50×50 square matrix. The multidimensional flight cost matrix constructed through this step provides input data for subsequent hierarchical clustering algorithms, enabling grid partitioning to simultaneously optimize flight efficiency and communication quality, avoiding the limitations of traditional methods that only consider geographical distance.

[0109] Using a multidimensional flight cost matrix as input, a hierarchical clustering algorithm is employed to aggregate all inspection points, resulting in an initial inspection grid. Hierarchical clustering is a bottom-up clustering method that gradually merges clusters with the highest similarity to ultimately form a hierarchical clustering structure.

[0110] In practice, the clustering state is first initialized, with each inspection point treated as an independent cluster. At this point, the number of clusters equals the total number of inspection points. Then, the aggregation step is repeated, with each iteration selecting the two closest clusters to merge, forming a new cluster. The distance between clusters is calculated using the average linking method, which calculates the average of the combined flight costs of all inspection point pairs in two clusters as the inter-cluster distance.

[0111] See Figure 2 Specifically, assuming cluster A contains m inspection points and cluster B contains n inspection points, the inter-cluster distance is the sum of the combined flight costs of all points in A and all points in B divided by m × n. Specifically, in the diagram, ∑∑ represents a double summation symbol, where the outer ∑ represents traversing all points in cluster A (i=1 to m), and the inner ∑ represents traversing all points in cluster B (j=1 to n). This represents a function for calculating the flight cost of a single point pair.

[0112] After each merging operation, the multidimensional flight cost matrix needs to be updated, and the inter-cluster distances containing the new clusters need to be recalculated. This iterative process continues until the number of clusters reaches a preset grid number threshold. This threshold needs to be set by comprehensively considering factors such as the area of ​​the inspection zone, the number of drones, and the complexity of the inspection task. For example, for an inspection zone covering 1000 square kilometers, equipped with 30 drones, the grid number threshold can be set to 5-8 to ensure that each grid has sufficient drone resources to perform inspection tasks. Each resulting cluster is defined as an initial inspection grid. Inspection points within each grid have high similarity in flight cost and communication quality, which is beneficial for drones to efficiently perform inspection tasks within the grid. The initial grid partitioning scheme formed through this step can well meet daily inspection needs under normal operating conditions.

[0113] During power grid operation, equipment fault alarm information is acquired in real time, and based on this information, the faulty power grid inspection points, at-risk equipment, and normal equipment are identified. Equipment fault alarm information originates from the power grid's monitoring and data acquisition system, which continuously monitors the operating status of various power equipment and automatically generates alarm information when an anomaly is detected. Alarm information typically includes key data such as the faulty equipment's unique identifier, geographical coordinates, fault occurrence time, and fault type.

[0114] In practice, the received fault alarm information is first parsed to extract the location information and fault type of the faulty equipment. Fault types include various situations such as line tripping, transformer overheating, insulator damage, and conductor strand breakage. Based on the fault type, the corresponding fault level is retrieved from a pre-established fault level mapping table. This table is a pre-built knowledge base that categorizes different types of faults into several levels according to severity. For example, a Level 1 fault indicates a serious impact on the safe operation of the power grid, requiring immediate action; a Level 2 fault indicates a potential chain reaction, requiring close monitoring; and a Level 3 fault indicates a minor anomaly, which can be addressed through routine inspections. Based on the retrieved fault level, the power grid inspection point where the fault occurred is determined. For higher fault levels, it is necessary not only to mark the inspection point where the faulty equipment is located but also to identify potentially affected inspection points in the surrounding area. Next, a risk assessment is conducted on all equipment within the inspection point where the fault occurred. The operating years and historical fault frequency data for each piece of equipment are obtained and stored in the equipment asset management database. A risk score is calculated for each piece of equipment, taking into account both the equipment's aging and fault history; the longer the operating years and the higher the historical fault frequency, the higher the risk score. The risk score is compared with a preset risk threshold. Equipment exceeding the threshold is defined as risky equipment, while equipment below the threshold is defined as normal equipment. This step enables dynamic classification of the power grid equipment status, providing accurate input data for subsequent construction of a fault potential field model.

[0115] Based on the equipment classification results determined in step S150, a fault potential energy model is constructed. The power grid inspection point where a fault occurs is defined as a potential energy source, and an initial potential energy value is set. The magnitude of the initial potential energy value is proportional to the fault level; the higher the fault level, the larger the initial potential energy value, reflecting the severity and urgency of the fault.

[0116] Risky devices are defined as low-potential wells, and corresponding potential energy values ​​are assigned. A low-potential well represents a region where potential energy easily accumulates. In this embodiment, risky devices, due to their inherent potential for failure, are more susceptible to the influence of surrounding faults and are therefore modeled as low-potential wells. The potential energy value of a low-potential well is set to a negative value, inversely proportional to the device's risk score; the higher the risk score, the larger the absolute value of the potential energy. For example, for a device with a risk score of 80, the potential energy value can be set to -30; for a device with a risk score of 60, the potential energy value can be set to -15. The potential energy of normal devices is set to 0, indicating that these devices are in a stable state, neither being a fault source nor exhibiting significant fault attraction characteristics. The potential energy model established through this step can quantitatively describe the fault state and risk level of different devices in the power grid, laying a theoretical foundation for subsequent simulation of the fault propagation process. The core idea of ​​this model is to transform abstract fault risk into a calculable potential energy value, enabling the complex fault evolution process to be modeled and predicted using mathematical methods.

[0117] The system acquires the electrical topology and geographical proximity of power grid inspection points and generates real-time fault potential field distribution maps based on these relationships to simulate the diffusion of potential energy from the source to the surrounding environment. Electrical topology describes the electrical connections between power equipment, such as the connection between substations and transmission lines, and the connection between transformers and busbars. These connections constitute the physical paths for power transmission and are also the main channels for fault propagation along electrical lines. Geographical proximity describes the spatial proximity of equipment; geographically close equipment may be affected by the same external environment, such as equipment in the same area potentially experiencing extreme weather events simultaneously.

[0118] In practical implementation, the power grid topology is first constructed. This graph is a weighted directed or undirected graph, where nodes represent power grid inspection points and edges represent electrical topology or geographical proximity relationships. For electrical topology relationships, the edge weights reflect the tightness of electrical connections, such as voltage levels and transmission capacities of transmission lines. For geographical proximity relationships, the edge weights reflect spatial distances. A diffusion coefficient is assigned to each edge of the power grid topology graph, which determines the rate at which potential energy propagates along that edge. The diffusion coefficient corresponding to electrical topology is usually greater than that corresponding to geographical proximity relationships because the speed and probability of fault propagation along electrical lines are generally higher than that through geographical space. For example, the diffusion coefficient for electrical topology can be set to 0.8, and the diffusion coefficient for geographical proximity relationships can be set to 0.3.

[0119] Based on the initial potential energy value and diffusion coefficient set in step S160, an iterative algorithm is used to calculate the potential energy value of each node in the power grid topology at each time step. The iterative process simulates the physical process of potential energy diffusion from a high-potential-energy region to a low-potential-energy region, similar to the numerical solution of heat conduction or diffusion equations. For any node, its potential energy value in the next time step is equal to its current potential energy value plus the potential energy increment transmitted from all neighboring nodes. The calculation of the potential energy increment considers the potential energy values ​​of neighboring nodes, the diffusion coefficient between nodes, and the potential energy gradient. Iteration stops when the number of iterations reaches a preset iteration threshold or the potential energy distribution converges.

[0120] The convergence criterion can be that the change in potential energy distribution between two consecutive iterations is less than a preset threshold, or the maximum potential energy gradient is less than a preset value. After obtaining the final potential energy distribution, it is mapped to a preset power grid geographic space to generate a real-time fault potential field distribution map. This distribution map is displayed in the form of a heat map or equipotential line map, intuitively showing the distribution of fault risk in the power grid space. High potential energy areas indicate areas with high fault risk that require key attention, while low potential energy areas indicate relatively safe areas. The real-time fault potential field distribution map generated through this step provides a crucial decision-making basis for subsequent grid dynamic adjustments.

[0121] Based on the real-time fault potential field distribution map, the total fault entropy value of the current power grid system is calculated. Then, a predefined set of grid operations is evaluated based on this total fault entropy value. The operation that maximizes the entropy deceleration rate is selected as the optimal decision, and the adjusted new grid boundary scheme is output. Fault entropy is used to quantify the dispersion of fault risk in the power grid space. A higher entropy value indicates a more dispersed and chaotic distribution of fault risk, and greater system uncertainty; a lower entropy value indicates a more concentrated distribution of fault risk, and a more ordered system state.

[0122] In practical implementation, the real-time fault potential field distribution map is first divided into multiple spatial units, each corresponding to a geographical grid. The potential energy value within each spatial unit is statistically analyzed, and the total potential energy of all spatial units is calculated. Then, the proportion of the potential energy value of each spatial unit to the total potential energy is calculated to obtain the potential energy distribution probability. Based on the potential energy distribution probability, the total fault entropy value of the current power grid system is calculated using the information entropy calculation formula, which is: ;in, represents the potential energy distribution probability of the i-th spatial unit; n represents the total number of spatial units.

[0123] Next, a predefined set of mesh operations is evaluated. This set includes operations such as merging adjacent meshes, splitting the current mesh, and exchanging parts of the mesh between meshes.

[0124] For each candidate operation, the grid state after the operation is simulated. The fault potential field distribution is recalculated based on the simulated grid state, and the estimated fault entropy value under the simulated state is calculated. The entropy deceleration rate corresponding to each candidate operation is calculated. The entropy deceleration rate is defined as the difference between the current total fault entropy value and the estimated fault entropy value divided by a preset time interval. This indicator reflects the rate at which the system entropy value decreases after the operation is executed. The candidate operation with the largest entropy deceleration rate is selected as the optimal decision because this operation can reduce the uncertainty of the system most quickly, making the fault risk distribution more concentrated and orderly, thereby improving the utilization efficiency of UAV inspection resources. The grid boundary is adjusted according to the optimal decision, and a new grid boundary scheme is output. For example, when a fault is detected to be rapidly spreading along a transmission line, crossing the original grid boundary, the optimal decision may be to merge multiple grids along the line to form a narrow emergency inspection grid, concentrating all UAV resources in the area to conduct intensive inspections of the fault zone. The dynamic grid adjustment mechanism implemented through this step can flexibly adjust the deployment of inspection resources according to the real-time evolution of the fault, effectively dealing with fault cascading propagation scenarios.

[0125] Based on the new grid boundary scheme, collaborative inspection tasks are generated and executed for each drone cluster within the new grid. The power grid drone inspection system includes a ground base station, which is connected to the drone cluster via a wireless communication network, responsible for issuing task commands and transmitting inspection data. In actual implementation, the new grid boundary scheme is first analyzed to obtain the boundary range of each new grid and the drone resources within the grid. The boundary range is represented by a geographic coordinate polygon, clearly defining the spatial area covered by the grid. Drone resource information includes status parameters such as the number of drones, model, current location, and remaining battery power. Based on the drone model and remaining battery power, the endurance of each drone is determined, i.e., the maximum distance it can fly with the current battery power. According to the real-time fault potential field distribution map, high-risk areas within each new grid where the potential energy value exceeds a preset potential energy threshold are identified.

[0126] High-risk areas are priority inspection zones, potentially containing faulty or high-risk equipment. The potential energy gradient of these high-risk areas is calculated, indicating the direction of the fastest increase in potential energy, which also indicates the fastest increase in fault risk. For each new grid cell, a collaborative inspection path is planned for the drone swarm within the new grid cell based on the potential energy gradient of the high-risk areas and the drone's endurance. Path planning employs optimization algorithms, such as genetic algorithms, particle swarm optimization, or improved traveling salesman problem algorithms. The planning objective is to maximize inspection coverage and efficiency while satisfying drone endurance constraints.

[0127] Specifically, inspection points are selected sequentially according to the priority inspection order determined by the potential energy gradient. For each inspection point, the flight distance from the current position of each UAV to that point is calculated. The flight distance is compared with the UAV's maximum endurance to filter candidate UAVs whose endurance meets the requirements. The UAV with the shortest flight distance is selected from the candidate UAVs, and the inspection point is assigned to that UAV, updating its remaining endurance. The path planning step is repeated until all inspection points are assigned or all UAVs' endurance is exhausted, generating a complete inspection path for each UAV in the UAV cluster. The collaborative inspection path is encapsulated as an inspection task instruction, which includes detailed information such as waypoint coordinate sequence, flight altitude, flight speed, hover time, and sensor parameters. The task instruction is sent to the corresponding UAV cluster via a ground base station. After receiving the instruction, the UAV automatically executes the inspection task, flying along the planned path and collecting inspection data. This step achieves a complete closed loop from grid partitioning to task execution, ensuring that UAV resources can quickly respond to fault conditions and efficiently complete inspection tasks.

[0128] This embodiment effectively solves the resource mismatch problem of traditional static grid partitioning in dealing with cascading fault propagation scenarios. By acquiring digital elevation model data and calculating the predicted flight distance and communication signal attenuation values ​​of the 3D geographic surface, the constructed multidimensional flight cost matrix fully considers the impact of terrain factors and communication quality on UAV inspection, making the initial grid partitioning more scientific and reasonable, and avoiding the limitations of traditional methods that only consider planar distances. The use of hierarchical clustering algorithm for grid partitioning can adaptively aggregate inspection points with similar flight costs and communication quality into the same grid, improving the collaborative efficiency within the grid. More importantly, by acquiring equipment fault alarm information in real time and constructing a fault potential energy model and a real-time fault potential field distribution map, the spatial distribution and evolution trend of fault risks can be intuitively predicted. Based on the dynamic grid adjustment mechanism of fault entropy, by evaluating the impact of different grid operations on the system entropy value, the operation that maximizes the entropy deceleration rate is selected as the optimal decision, realizing intelligent dynamic adjustment of the grid boundary. This allows UAV resources to be flexibly redeployed as the fault evolves, effectively avoiding the waste and shortage of emergency resources. Based on the adjusted new grid boundary scheme and combined with potential energy gradient information, collaborative inspection tasks are generated for the UAV swarm, ensuring that high-risk areas receive priority and intensive inspection coverage. This significantly improves the timeliness of fault detection and handling, thereby increasing inspection efficiency. It is not only suitable for emergency scenarios such as extreme weather or cascading failures, but also enables dynamic optimization of inspection resource allocation based on equipment status during daily operation, comprehensively enhancing the intelligence level and emergency response capabilities of the power grid UAV inspection system.

[0129] In one embodiment of this invention, a multidimensional flight cost matrix is ​​constructed based on the predicted flight distance and communication signal attenuation values ​​of a three-dimensional geographic surface, including the following steps:

[0130] S210. For any pair of inspection points, multiply the flight distance of the three-dimensional geographic surface by the preset flight distance weighting coefficient to obtain the first weighted value, and multiply the predicted value of communication signal attenuation by the preset communication attenuation weighting coefficient to obtain the second weighted value.

[0131] S220. Sum the first weighted value and the second weighted value to obtain the comprehensive flight cost of the inspection point pair;

[0132] S230. Organize the comprehensive flight costs of all inspection point pairs into a matrix form to obtain a multidimensional flight cost matrix.

[0133] In practice, for any pair of inspection points, the predicted flight distance and communication signal attenuation on the 3D geographic surface need to be converted into a unified cost metric. Since flight distance and communication attenuation have different dimensions and numerical ranges, they cannot be directly compared and synthesized; therefore, weighted processing is required to achieve normalization.

[0134] First, the flight distance on the 3D geographic surface is multiplied by a preset flight distance weighting coefficient to obtain the first weighted value. The flight distance weighting coefficient reflects the importance of flight distance in the total cost assessment. The setting of this coefficient needs to comprehensively consider the UAV's energy consumption characteristics, the timeliness requirements of the inspection mission, and the geographical characteristics of the power grid inspection area. For example, in flat plains with good communication conditions, flight distance is the main factor affecting inspection efficiency, and the flight distance weighting coefficient can be set to 0.7; while in mountainous areas with complex terrain and limited communication, the importance of communication quality is more prominent, and the flight distance weighting coefficient can be appropriately reduced to 0.4.

[0135] In the specific calculation, assuming the flight distance between inspection point A and inspection point B on the three-dimensional geographic surface is 5.2 kilometers, and the flight distance weighting coefficient is set to 0.6, then the first weighted value is 3.12. This weighted value quantifies the impact of flight distance on the drone's performance of inspection tasks. The longer the flight distance, the more power the drone consumes, the longer the task execution time, and the higher the corresponding cost.

[0136] Simultaneously, the predicted communication signal attenuation value is multiplied by a preset communication attenuation weighting coefficient to obtain the second weighted value. The communication attenuation weighting coefficient reflects the importance of communication quality in inspection tasks, especially in scenarios requiring real-time transmission of high-definition video or large amounts of sensor data, where a stable communication link is crucial. The predicted communication attenuation value is usually expressed in decibels (dB); a higher value indicates more severe signal attenuation and poorer communication quality. For example, if the predicted communication signal attenuation between inspection point A and inspection point B is 85 dB, and the communication attenuation weighting coefficient is set to 0.4, then the second weighted value is 34.

[0137] To ensure the comparability of the two weighted values ​​across their numerical ranges, the predicted communication attenuation needs to be normalized in practical applications, such as mapping it to a numerical range close to the flight distance. This weighting process transforms the previously incomparable flight distance and communication attenuation into a unified cost metric.

[0138] After obtaining the first and second weighted values, the two are summed to obtain the comprehensive flight cost of the inspection point pair. The comprehensive flight cost takes into account both the physical and communication costs of the UAV flying between the two inspection points, and can more comprehensively reflect the correlation strength between the inspection points. Continuing the previous example, with the first weighted value of 3.12 and the second weighted value of 34, the comprehensive flight cost is 37.12.

[0139] The lower the overall flight cost, the shorter the flight distance between the two inspection points and the better the communication quality. This indicates higher efficiency for the drone in performing tasks between the two points. Therefore, these two inspection points are more suitable to be assigned to the same grid, facilitating efficient collaborative inspections by the drone within the grid. Conversely, a higher overall flight cost indicates significant flight obstacles or communication difficulties between the two inspection points. Assigning them to the same grid may lead to a waste of drone resources or a reduction in task execution efficiency.

[0140] In practical applications, the calculation of comprehensive flight costs can be further optimized. For example, a nonlinear weighting function can be introduced to significantly increase the weight of communication costs when communication attenuation exceeds a certain critical threshold, so as to avoid classifying inspection points with extremely poor communication conditions into the same grid.

[0141] After calculating the comprehensive flight cost of all inspection point pairs, these cost values ​​are organized into a matrix to obtain a multidimensional flight cost matrix. This matrix is ​​the core input data for subsequent hierarchical clustering algorithms, and its structure and quality directly affect the grid division effect.

[0142] In practice, the multidimensional flight cost matrix is ​​a square matrix whose number of rows and columns equals the total number of inspection points within the inspection area. Assuming the inspection area contains N inspection points, the multidimensional flight cost matrix is ​​an N×N square matrix. The element in the i-th row and j-th column represents the combined flight cost between inspection point i and inspection point j. Since the calculation of the combined flight cost is symmetric—that is, the cost from inspection point A to inspection point B is the same as the cost from inspection point B to inspection point A—the multidimensional flight cost matrix is ​​a symmetric matrix. The diagonal elements of the matrix represent the cost of an inspection point traveling to itself, which is usually set to 0.

[0143] In constructing the matrix, it is necessary to traverse all possible combinations of inspection point pairs. For N inspection points, there are a total of For each different inspection point pair, a corresponding number of comprehensive flight costs need to be calculated. For example, for an inspection area containing 50 inspection points, the comprehensive flight costs of 1225 inspection point pairs need to be calculated, and these cost values ​​need to be filled into a 50×50 matrix.

[0144] Once the matrix is ​​constructed, matrix operations and data analysis methods can be used to delve deeper into the relationships between inspection points. For example, eigenvalues ​​and eigenvectors of the matrix can be calculated to identify key nodes and connections within the inspection area; cluster analysis can be performed on the matrix to preliminarily identify groups of inspection points with similar cost characteristics.

[0145] The construction of the multidimensional flight cost matrix transforms complex geospatial and communication quality relationships into a structured numerical matrix, providing a standardized input interface for subsequent algorithm processing. This greatly simplifies the implementation complexity of the grid partitioning algorithm and also facilitates the visualization and analysis of the grid partitioning results.

[0146] This embodiment constructs a multidimensional flight cost matrix by introducing a weighting mechanism, effectively solving the problem of the inability to directly compare the two heterogeneous indicators of flight distance and communication quality. The first weighted value is obtained by multiplying the flight distance on the three-dimensional geographic surface by the flight distance weighting coefficient, and the second weighted value is obtained by multiplying the predicted communication signal attenuation value by the communication attenuation weighting coefficient, thus achieving normalization of indicators with different dimensions. The comprehensive flight cost obtained by summing the two weighted values ​​can comprehensively reflect the real cost of the UAV performing tasks between inspection points, considering both flight energy consumption and time costs, as well as the impact of communication quality on data transmission. The flexible setting mechanism of the weighting coefficients allows this method to adapt to the needs of different application scenarios. In plain areas with good communication conditions, the flight distance weight can be increased, while in mountainous areas with complex terrain, the communication quality weight can be increased, enhancing the versatility and adaptability of the method. Organizing the comprehensive flight cost of all inspection point pairs into a matrix form provides a standardized input interface for subsequent hierarchical clustering algorithms, simplifying the complexity of algorithm implementation, improving computational efficiency, and effectively enhancing the quality of the initial grid division.

[0147] In one embodiment of this invention, the multidimensional flight cost matrix is ​​used as input to aggregate all inspection points to obtain an initial inspection grid, including the following steps:

[0148] S310. Initialize the clustering state, treating each inspection point as an independent cluster;

[0149] S320. Repeat the aggregation step until the number of clusters reaches the preset grid number threshold.

[0150] S330. Define each cluster obtained at the end as the initial inspection grid;

[0151] The polymerization step includes:

[0152] Calculate the distance between any two clusters based on the multidimensional flight cost matrix;

[0153] The two clusters with the smallest distance are selected and merged to form a new cluster;

[0154] Update the multidimensional flight cost matrix, recalculate the inter-cluster distances that include the new clusters, and determine the number of current clusters based on the inter-cluster distances.

[0155] In practice, the first step of the hierarchical clustering algorithm is to initialize the clustering state, treating each inspection point as an independent cluster. This bottom-up initialization strategy ensures that the algorithm can start from the finest-grained division, gradually merge similar inspection points, and finally form a reasonable grid structure.

[0156] Assume the inspection area contains N inspection points, labeled P1, P2, P3, ..., PN. During the initialization phase, each inspection point Pi is considered an independent cluster Ci. At this point, the total number of clusters equals the total number of inspection points, i.e., the initial number of clusters is N. Each cluster initially contains only one inspection point, and the center of the cluster is the geographical coordinate of that inspection point. The cluster's attribute information includes all characteristic parameters of that inspection point, such as altitude, equipment type, and distance to communication base stations.

[0157] To facilitate subsequent clustering operations, an index structure needs to be established for each cluster, recording the cluster's unique identifier, the list of inspection points it contains, the cluster's center coordinates, and its distance information from other clusters. For example, for an inspection area containing 50 inspection points, 50 independent clusters will be formed after initialization, with each cluster containing one inspection point. The initialization process also requires preparing an inter-cluster distance matrix. The initial state of this matrix is ​​the same as the multidimensional flight cost matrix, because at this point each cluster contains only one inspection point, and the inter-cluster distance is equal to the combined flight cost between the corresponding inspection points.

[0158] After initialization, the aggregation step is repeated until the number of clusters reaches the preset grid size threshold. The aggregation step is the core iterative process of the hierarchical clustering algorithm. By continuously merging the nearest clusters, the number of clusters is gradually reduced, ultimately forming a grid partitioning scheme that meets the requirements.

[0159] The preset grid number threshold is a key parameter determined based on actual application needs, requiring comprehensive consideration of multiple factors. First, the area and complexity of the inspection area must be considered; larger areas and more complex terrains necessitate a greater number of grids to ensure a moderate workload within each grid. Second, the availability of drone resources must be considered; the number of grids should match the number of drones to ensure sufficient drones for inspection tasks within each grid. For example, for an inspection system equipped with 30 drones, the grid number threshold can be set to 6, allocating an average of 5 drones per grid. Furthermore, the response time requirements and collaborative efficiency of the inspection tasks must also be considered.

[0160] The aggregation step first calculates the distance between any two clusters based on the multidimensional flight cost matrix. Initially, since each cluster contains only one checkpoint, the inter-cluster distance is directly read from the multidimensional flight cost matrix. As the clustering process progresses, some clusters will contain multiple checkpoints, at which point a specific distance calculation method is needed to determine the inter-cluster distance.

[0161] In this embodiment, the average linking method is used. This method calculates the average of the comprehensive flight costs between all pairs of inspection points in two clusters as the inter-cluster distance. Specifically, assuming cluster A contains m inspection points and cluster B contains n inspection points, the inter-cluster distance is equal to the sum of the comprehensive flight costs of all points in A and all points in B divided by... This method can effectively balance the compactness and segregation of clusters. For example, if cluster A contains checkpoints P1 and P2, and cluster B contains checkpoint P3, then the inter-cluster distance is... Where Cost represents the total flight cost read from the multidimensional flight cost matrix.

[0162] Next, the two clusters with the smallest distance are selected and merged to form a new cluster. After calculating the distances between all cluster pairs, the inter-cluster distance matrix needs to be traversed to find the cluster pair with the smallest distance. Assuming that clusters Ci and Cj have the smallest distance, these two clusters are merged into a new cluster Cnew. The new cluster contains all the checkpoints from the original two clusters; that is, the set of checkpoints in Cnew is the union of Ci and Cj.

[0163] After the merge operation is completed, the original clusters Ci and Cj are deleted from the cluster list, and the new cluster Cnew is added to the cluster list, reducing the total number of clusters by 1. For example, in a certain iteration, the distance between cluster C1 (containing P1 and P2) and cluster C5 (containing P5) is the smallest, which is 25.3. Then they are merged into a new cluster C15 (containing P1, P2, and P5), reducing the total number of clusters from the current 10 to 9.

[0164] Then, the multidimensional flight cost matrix is ​​updated, and the inter-cluster distances, including the new clusters, are recalculated. Since the merging operation generates new clusters, the distances between the new clusters and all other existing clusters need to be calculated, and the inter-cluster distance matrix is ​​updated. For the new cluster Cnew and any other cluster Ck, the average linkage method is used to calculate their distances, i.e., the average of the combined flight costs between all checkpoints in Cnew and all checkpoints in Ck. The updated inter-cluster distance matrix has one less dimension.

[0165] Based on the updated inter-cluster distance, the number of current clusters is determined, and it is checked whether a preset grid number threshold has been reached. If the current number of clusters is greater than the threshold, the next round of aggregation is executed; if it is equal to the threshold, the iteration stops. Through this iterative process, the number of clusters gradually decreases, and the optimal merging scheme is selected in each iteration, ensuring that the final grid partitioning scheme achieves an optimal balance between flight cost and communication quality.

[0166] After the aggregation iteration process is completed, each cluster obtained is defined as the initial inspection grid. At this point, the number of clusters has reached the preset grid number threshold, and each cluster represents an independent inspection grid, containing several inspection points that are geographically close and have similar flight costs and communication quality.

[0167] In practical implementation, post-processing is required for each cluster to extract and record key attribute information of the grid. First, a unique identifier is assigned to each grid for easier subsequent management and scheduling. Second, the geographical boundary of each grid is calculated by performing convex hull calculation or buffer analysis on the coordinates of all inspection points within the grid to obtain the polygonal boundary of the grid. Determining the grid boundary is crucial for restricting the flight area of ​​UAVs and for cross-grid coordination.

[0168] Secondly, statistical information such as the number of inspection points, the distribution of inspection point types, and the average altitude within each grid is collected to provide a basis for subsequent resource allocation. In addition, the center coordinates of the grid need to be calculated as the representative location of that grid; this can be achieved using the geometric center or weighted center of the coordinates of all inspection points within the grid. For example, a grid containing 8 inspection points distributed along a transmission line, with the grid boundary being a narrow, elongated area extending along the line, and the grid center located at the midpoint of the line.

[0169] This embodiment achieves intelligent grouping of inspection points through a hierarchical clustering algorithm, effectively improving the scientificity and rationality of grid partitioning. During initial clustering, each inspection point is treated as an independent cluster, employing a bottom-up aggregation strategy to ensure the algorithm systematically explores all possible cluster combinations starting from the finest granularity. The iterative process of aggregation is repeated, merging the two clusters with the smallest distance each time, adhering to the principle of local optima and gradually forming a globally optimal grid partitioning scheme. The average linkage method is used to calculate inter-cluster distances, effectively balancing the compactness and segregation of clusters and improving the quality of clustering results. By setting a grid number threshold as the iteration termination condition, the final grid number matches the available UAV resources, ensuring sufficient UAVs to perform inspection tasks within each grid. A dynamic update mechanism that updates the multidimensional flight cost matrix and recalculates inter-cluster distances ensures that each merging operation is based on the latest clustering state, improving the algorithm's accuracy. The resulting initial inspection grid can effectively support the daily inspection tasks of UAVs under normal operating conditions. The inspection points within the grid have high correlation, reducing the frequency of cross-grid flights and communication switching, and significantly improving the overall inspection efficiency and resource utilization.

[0170] In one embodiment of this invention, determining the faulty power grid inspection point, at-risk equipment, and normal equipment based on equipment fault alarm information includes the following steps:

[0171] S410. Analyze the equipment fault alarm information to obtain the location information and fault type of the faulty equipment;

[0172] S420. Query the fault level from the preset fault level mapping table according to the fault type, and determine the power grid inspection point where the fault occurred based on the fault level.

[0173] S430. Obtain the operating years and historical fault frequency of each device at the power grid inspection point where a fault occurred, and calculate the risk score of each device.

[0174] S440. Devices with risk scores exceeding a preset risk threshold are identified as risky devices, and devices with risk scores not exceeding the preset risk threshold are identified as normal devices.

[0175] In practice, the power grid monitoring and data acquisition system continuously monitors the operating status of various power equipment. When an equipment anomaly is detected, it automatically generates fault alarm information. Fault alarm information is usually transmitted in the form of structured data packets, containing several key fields, such as the equipment's unique identifier, equipment name, name of the substation or line to which it belongs, geographical coordinates, fault occurrence timestamp, fault type code, fault description text, and alarm level.

[0176] The parsing process first requires receiving and reading alarm information data packets, and extracting the content of each field according to a predefined data format protocol. For location information extraction, latitude and longitude data can be directly obtained from the geographic coordinate field, such as 118.5 degrees east longitude and 32.3 degrees north latitude. If the alarm information does not directly contain coordinates, the device asset management database needs to be queried using the device identification code to obtain the installation location coordinates of the device. The accuracy of the location information is crucial for subsequently determining the fault inspection point; therefore, coordinate validity verification is required during the parsing process to ensure that the coordinate values ​​are within a reasonable range and located within the inspection area.

[0177] For fault type extraction, alarm information typically includes a fault type code. The parsing program needs to identify the fault type code and retrieve the corresponding fault type name and detailed description from a pre-defined fault type dictionary.

[0178] In some cases, alarm messages may contain multiple fault types, indicating that the device has experienced multiple anomalies simultaneously. In such cases, it is necessary to extract all fault types, prioritize them, and select the most severe fault type as the primary fault type. This parsing operation transforms the raw alarm information into structured fault data.

[0179] After obtaining the fault type, it is necessary to query the fault level from a pre-defined fault level mapping table based on the fault type, and then determine the power grid inspection point where the fault occurred based on the fault level. The fault level mapping table is a pre-established knowledge base that classifies various fault types into several levels according to their degree of impact on the safe operation of the power grid.

[0180] In practice, fault levels are typically divided into three levels. A Level 1 fault indicates a serious threat to power grid safety, potentially leading to widespread power outages or equipment damage, requiring immediate attention. A Level 2 fault indicates a potential chain reaction or impact on power supply to critical users, requiring urgent monitoring. A Level 3 fault indicates a minor anomaly that can be addressed through routine inspections. The fault level mapping table is stored in tabular form, with each row containing information such as fault type code, fault type name, fault level, scope of impact, and suggested response time.

[0181] The query process uses the fault type code as an index key to quickly locate the corresponding fault level record in the mapping table. After obtaining the fault level, it is necessary to determine the range of power grid inspection points where the fault occurred based on the fault level. For a Level 1 fault, it is necessary not only to mark the inspection point where the faulty equipment is located, but also to identify adjacent inspection points directly connected to the equipment in the electrical topology, because a Level 1 fault can propagate rapidly to adjacent equipment through electrical lines. For example, if a 220kV transmission line trips, and this line connects substation A and substation B, then the inspection points on the line itself, as well as the inspection points on both substation A and substation B, need to be marked as the inspection points where the fault occurred.

[0182] For Level 2 faults, the inspection point where the faulty equipment is located and its geographically nearest inspection points are marked, resulting in a relatively small area. For Level 3 faults, only the single inspection point where the faulty equipment is located is marked. Through this level determination and area identification, the precise identification of the fault's impact range is achieved.

[0183] After identifying the power grid inspection points where the fault occurred, a risk assessment needs to be conducted on all equipment within these inspection points. First, the operational lifespan and historical fault frequency data for each piece of equipment are obtained. This data is stored in the equipment asset management system and the maintenance history database. Operational lifespan refers to the time span from the equipment's commissioning date to the current date, calculated in years. Operational lifespan reflects the degree of equipment aging; generally, the longer the operational lifespan, the more likely key parameters such as insulation performance and mechanical strength are to degrade, and the higher the risk of failure.

[0184] Historical fault frequency refers to the number of times equipment has failed within a certain period, typically based on fault records from the last 3 or 5 years. For example, if a transmission line has experienced 2 tripping faults and 1 lightning strike fault in the past 3 years, its historical fault frequency is 3. Historical fault frequency reflects the reliability level of the equipment; a higher frequency indicates that the equipment is more prone to problems.

[0185] After obtaining these two data points, a risk score is calculated for each device. The risk score uses a weighted comprehensive scoring method, which sums the results after multiplying the years of operation and the frequency of historical failures by their respective weighting coefficients. The weighting coefficient for years of operation is typically set to 0.6, and the weighting coefficient for the frequency of historical failures is set to 0.4, reflecting the dominant role of aging factors in risk assessment.

[0186] To make the two indicators comparable, the raw data needs to be normalized. The number of years of operation can be divided by the equipment's design life to obtain a normalized value. For example, if a transformer's design life is 30 years, the normalized value after 15 years of operation is 0.5. The historical failure frequency can be divided by the maximum possible number of failures within the statistical period to obtain a normalized value. The final risk score is calculated as follows: Risk score equals the normalized value of the number of years of operation multiplied by 0.6, plus the normalized value of the historical failure frequency multiplied by 0.4, multiplied by 100 to obtain a percentage score.

[0187] After calculating the risk score for each device, the devices need to be categorized into two types based on a preset risk threshold: high-risk devices and normal devices. The risk threshold is a critical value determined based on historical data statistical analysis and expert experience, used to distinguish between high-risk and low-risk devices. In practice, the risk threshold is typically set at 60 points; devices with a risk score exceeding 60 are considered to have a higher risk of failure and require close monitoring and preventative maintenance.

[0188] Setting a risk threshold requires striking a balance between false alarm rate and false negative rate. A threshold that is too low can lead to many normal devices being mistakenly identified as risky, increasing unnecessary inspection workload; a threshold that is too high may miss truly high-risk devices, creating potential safety hazards. By statistically analyzing historical fault data, a curve showing the relationship between risk score and actual fault occurrence rate can be plotted, and the score value at the inflection point of the curve can be selected as the risk threshold.

[0189] Each device's risk score is compared to a risk threshold. If the risk score is 60 or higher, the device is identified as a high-risk device, marked as high-risk in the system, and added to the risk device list. If the risk score is less than 60, the device is identified as a normal device and remains under normal monitoring. For example, at a faulty inspection point containing 5 devices, transformer A has a risk score of 75, circuit breaker B has a risk score of 68, surge arrester C has a risk score of 45, current transformer D has a risk score of 38, and disconnector E has a risk score of 52. In this case, transformer A and circuit breaker B are identified as high-risk devices, while the other three devices are normal devices.

[0190] This embodiment achieves accurate identification and classification of power grid equipment fault risks through a multi-level equipment status assessment mechanism. The process of parsing equipment fault alarm information and extracting location information and fault type transforms raw alarm data into structured fault data, improving data processing efficiency and accuracy. By querying the fault level mapping table based on the fault type and determining the inspection point range where the fault occurred, rapid assessment of the fault's impact range is achieved. For high-level faults, potentially affected adjacent equipment can be identified in a timely manner, avoiding missed fault range assessments. The method of obtaining equipment operating years and historical fault frequencies and calculating risk scores comprehensively considers two key factors: equipment aging and reliability, enabling objective quantification of equipment fault risk levels. Through normalization processing and weighted comprehensive scoring, equipment of different types and operating states becomes comparable, improving the scientific rigor of risk assessment. Comparing risk scores with preset thresholds and classifying equipment into risky and normal equipment enables dynamic risk-layered management of equipment. This allows high-risk equipment to receive focused attention and preventative maintenance, providing reliable decision support for dynamic grid adjustments and emergency inspections.

[0191] In one embodiment of this invention, generating a real-time fault potential field distribution map based on electrical topology and geographical proximity includes the following steps:

[0192] S510. Construct a power grid topology graph, where nodes in the power grid topology graph represent power grid inspection points, and edges represent electrical topology or geographical proximity relationships.

[0193] S520. Assign a diffusion coefficient to each edge of the power grid topology graph, wherein the diffusion coefficient corresponding to the electrical topology is greater than the diffusion coefficient corresponding to the geographical proximity relationship.

[0194] S530. Based on the initial potential energy value and diffusion coefficient, calculate the potential energy value of each node in the power grid topology at each time step.

[0195] S540. Stop iterating when the number of iterations reaches the preset iteration threshold or the potential energy distribution converges, and obtain the final potential energy distribution.

[0196] S550: Map the final potential energy distribution to the preset power grid geographic space to generate a real-time fault potential field distribution map.

[0197] In practical implementation, constructing a power grid topology map is a fundamental step in simulating fault propagation. A power grid topology map is a complex network structure that comprehensively reflects the electrical connections and spatial proximity relationships between power grid equipment. Nodes in the map represent power grid inspection points, each containing a unique identifier, geographical coordinates, a list of included devices, and current potential energy values. Edges represent the relationships between inspection points and are divided into two types: electrical topology edges and geographical proximity edges.

[0198] Electrical topology edges reflect the physical connections between electrical equipment, such as transmission lines connecting two substations or transformers connecting busbars of different voltage levels. These connections constitute the actual paths for power transmission and are also the main channels for fault propagation along electrical lines. When constructing electrical topology edges, connection relationship data needs to be extracted from the primary wiring diagram and equipment ledger of the power grid. For example, if 220kV substation A is connected to substation B via a transmission line, then an electrical topology edge is established between node A and node B in the topology diagram. The edge attributes include parameters such as line length, voltage level, and transmission capacity.

[0199] Geographic proximity edges reflect the spatial proximity of inspection points. Even if two inspection points are not directly electrically connected, if they are geographically adjacent, they may fail simultaneously due to shared external factors such as extreme weather or geological disasters. When constructing geographic proximity edges, a distance threshold method is used to calculate the geographical distance between any two inspection points. If the distance is less than a preset proximity threshold, a geographic proximity edge is established. The proximity threshold needs to be determined based on the actual situation of the inspection area; it can be set to 2 kilometers in urban power grids and 5 kilometers in rural power grids. For example, if substation C and substation D are not directly electrically connected but are only 1.5 kilometers apart, a geographic proximity edge is established between nodes C and D in the topology graph.

[0200] After constructing the power grid topology, a diffusion coefficient needs to be assigned to each edge. This coefficient determines the rate and intensity of potential energy propagation along that edge. The setting of the diffusion coefficient reflects the difference in the impact of different propagation paths on fault propagation. In practice, the diffusion coefficient corresponding to electrical topology is significantly larger than that corresponding to geographical proximity, because the speed and probability of fault propagation along electrical lines are much higher than that through geographical space.

[0201] The diffusion coefficient of an electrical topology is typically set between 0.6 and 0.9, with the specific value determined based on the tightness and importance of the electrical connections. For substations connected to high-voltage transmission lines, a diffusion coefficient of 0.85 can be used, indicating a high probability of fault propagation along the line. For equipment connected to low-voltage distribution lines, a diffusion coefficient of 0.65 can be used, reflecting a lower propagation risk. Determining the diffusion coefficient also requires consideration of dynamic factors such as line load rate and equipment health status. For example, for heavily loaded lines with a load rate exceeding 80%, a correction value of 0.1 can be added to the base diffusion coefficient, as equipment is more prone to cascading failures under heavy load conditions.

[0202] The diffusion coefficient for geographical proximity is typically set between 0.2 and 0.4, reflecting the correlation of failures caused by external environmental factors. For inspection points located in the same geologically hazardous area, the diffusion coefficient can be set to 0.35; for inspection points that are only spatially close but have significantly different environmental conditions, the diffusion coefficient can be set to 0.25. In extreme weather scenarios, such as during typhoons or freezing disasters, the diffusion coefficient for geographical proximity needs to be dynamically adjusted and can be temporarily increased to 0.5 to 0.6 to reflect the simultaneous impact of severe weather on a large area of ​​equipment.

[0203] Based on the initial potential energy value and diffusion coefficient, an iterative algorithm is used to calculate the potential energy value of each node in the power grid topology at each time step, simulating the dynamic process of potential energy diffusion from high potential energy regions to low potential energy regions. In practical implementation, the iterative calculation process is similar to the numerical solution of the heat conduction equation or diffusion equation.

[0204] For any node i, first obtain the set of all neighboring nodes adjacent to node i. A neighboring node is a node in the topology graph that is directly connected to node i through an edge. Then, calculate the difference between the potential energy value of each neighboring node j at the current time step t and the potential energy value of node i at the current time step t, denoted as the potential energy gradient. The sign of the potential energy gradient indicates the direction of potential energy propagation; a positive value indicates that potential energy flows from neighboring node j to node i, and a negative value indicates that potential energy flows from node i to neighboring node j.

[0205] Multiplying the potential gradient by the diffusion coefficient of the edge connecting nodes i and j yields the potential energy transfer from that neighboring node to node i. This potential energy transfer reflects how much potential energy is transferred to node i through that edge within a given time step. For all neighboring nodes of node i, the potential energy transfer is calculated separately, and then all transfer amounts are summed to obtain the total potential energy increment received by node i from all neighboring nodes at the current time step. The potential energy value of node i at the next time step t+1 is equal to its potential energy value at the current time step t plus the total potential energy increment. This can be expressed by the formula:

[0206] ;

[0207] in, This represents the potential energy value of node i at time step t; Represents the set of neighboring nodes of node i; This represents the diffusion coefficient of the edge connecting node i and node j. This formula reflects the physical law that potential energy diffuses along the direction of the potential energy gradient; the larger the diffusion coefficient, the larger the potential energy gradient, and the faster the potential energy is transferred.

[0208] During the iteration process, special handling of potential energy sources and low potential wells also needs to be considered. For nodes defined as potential energy sources, their potential energy value remains at the initially set high value during the iteration process, continuously outputting potential energy to surrounding nodes to simulate the continuous impact of the fault source. For nodes defined as low potential wells, since their initial potential energy value is negative, they will attract the potential energy of surrounding nodes to accumulate towards them, simulating the characteristic that risky equipment is more susceptible to the impact of faults.

[0209] The iterative calculation process requires setting reasonable termination conditions to ensure both a stable potential energy distribution and avoid unnecessary computational overhead. In practice, two termination conditions are used: the number of iterations reaches a preset iteration threshold or the potential energy distribution converges. The preset iteration threshold is the maximum number of iterations estimated based on the grid size and diffusion rate, typically set to 50 to 200. For a medium-sized grid with 100 inspection points, the iteration threshold can be set to 100 to ensure sufficient time for the potential energy to diffuse from the fault source to the entire network.

[0210] The criterion for determining convergence of the potential energy distribution is that the change in potential energy distribution between two consecutive iterations is less than a preset convergence threshold. Specifically, the sum of the squares of the differences in potential energy values ​​at all nodes between time step t+1 and time step t is calculated. If this value is less than the convergence threshold, the potential energy distribution is considered to have reached a steady state, and further iterations are unnecessary. The convergence threshold is typically set between 0.01 and 0.001; a smaller value indicates a higher requirement for convergence accuracy.

[0211] After each iteration, the system first checks if the number of iterations has reached a threshold. If it has, the iteration stops immediately. If not, the change in potential energy distribution is calculated to determine if the convergence condition is met. When any termination condition is met, the iteration stops and the final potential energy distribution is obtained. The final potential energy distribution is a vector containing the potential energy values ​​of all nodes in the power grid topology diagram. This distribution reflects the fault risk level of each inspection point in the power grid under the current fault condition.

[0212] After obtaining the final potential energy distribution, it needs to be mapped to a preset power grid geographic space to generate a real-time fault potential field distribution map, which visually displays the spatial distribution of fault risks. In actual implementation, a rasterized representation of the power grid geographic space is first established, dividing the entire inspection area into regular grid cells, each cell corresponding to a geographic coordinate range. The size of the grid cell is determined based on the area and accuracy requirements of the inspection area. For an area covering 1000 square kilometers, the grid cell can be set to 500 meters × 500 meters, forming approximately 4000 grid cells.

[0213] Then, the potential energy value of each node in the power grid topology is mapped to its corresponding geographical coordinates. If a node is located exactly at the center of a grid cell, its potential energy value is directly assigned to that grid cell. If a node is located at the boundary of a grid cell or multiple nodes are located in the same grid cell, a weighted average or maximum value selection method is used to determine the potential energy value of the grid cell. For grid cells without inspection points, spatial interpolation methods are used to estimate their potential energy value. Commonly used interpolation methods include inverse distance weighted interpolation and Kriging interpolation. The interpolation process considers the potential energy values ​​and distances of surrounding inspection points; the closer the inspection point, the greater its contribution to the potential energy of the grid cell.

[0214] After calculating the potential energy values ​​of all grid cells, a real-time fault potential field distribution map is generated using visualization technology. The distribution map is usually displayed in the form of a heat map or equipotential line map, using different colors to represent different potential energy levels, such as red for high potential energy regions, yellow for medium potential energy regions, and green for low potential energy regions.

[0215] This embodiment constructs a power grid topology map and simulates the potential energy diffusion process, applying the potential energy theory from physics to power grid fault propagation modeling, thus achieving an intuitive prediction of the spatial distribution and evolution trend of fault risk. The constructed power grid topology map, incorporating electrical topology and geographical proximity, comprehensively describes the network structure of the power grid, reflecting both the main channels of fault propagation along electrical lines and considering spatial correlations caused by external environmental factors, thereby improving the model's comprehensiveness and accuracy. Differential diffusion coefficients are assigned to different types of edges, reflecting the different roles of electrical topology and geographical proximity in fault propagation, ensuring that the simulation results realistically reflect the propagation characteristics of faults in the power grid. The iterative calculation process based on the initial potential energy value and diffusion coefficient simulates the dynamic process of potential energy diffusion from high-potential-energy regions to low-potential-energy regions, revealing the spatiotemporal evolution law of fault risk. Setting dual termination conditions—an iteration threshold and a convergence judgment—ensures that the potential energy distribution reaches a steady state while avoiding unnecessary computational overhead, improving the algorithm's efficiency. The final potential energy distribution is mapped to the real-time fault potential field distribution map generated by the power grid geospatial model. The spatial distribution of fault risk is displayed intuitively in the form of a heat map, providing dispatchers with a visual decision-making tool, which facilitates the rapid identification of high-risk areas and effectively improves the pertinence and timeliness of emergency inspections.

[0216] In one embodiment of this example, the potential energy value of each node in the power grid topology at each time step is calculated based on the initial potential energy value and the diffusion coefficient, including the following steps:

[0217] S610. For any node, obtain all neighboring nodes adjacent to that node;

[0218] S620. Calculate the difference between the potential energy value of the neighboring node at the current time step and the potential energy value of this node at the current time step.

[0219] S630. Multiply the difference in potential energy values ​​by the diffusion coefficient of the corresponding edge to obtain the potential energy transfer amount.

[0220] S640. Sum the potential energy transferred from all neighboring nodes to this node, and add the potential energy value of this node at the current time step to obtain the potential energy value of this node at the next time step.

[0221] In practical implementation, for any node in the power grid topology, it is first necessary to obtain all neighboring nodes adjacent to that node. This is a prerequisite step for calculating potential energy diffusion. Neighboring nodes are nodes in the topology that are directly connected to the target node via edges. These connections may be electrical topology edges or geographically proximate edges. Obtaining neighboring nodes is achieved by traversing the adjacency list or adjacency matrix of the topology graph.

[0222] In the adjacency list representation, each node maintains a list recording the identifiers of all nodes connected to it and their corresponding edge attributes. For example, for node A, its adjacency list might contain nodes B, C, and D, indicating that node A is directly connected to these three nodes. In the adjacency matrix representation, the element in the i-th row and j-th column indicates whether there is a connection between node i and node j. If there is, the element value is the diffusion coefficient of the corresponding edge; otherwise, it is 0. By querying the i-th row of the adjacency matrix, all neighboring nodes of node i can be quickly found.

[0223] When acquiring neighboring nodes, it is also necessary to simultaneously obtain the attribute information of the connecting edges, especially the diffusion coefficient, as this coefficient will be used in subsequent potential energy transfer calculations. For example, node A's neighbors include node B (connected via an electrical topology edge, diffusion coefficient 0.85), node C (connected via an electrical topology edge, diffusion coefficient 0.75), and node D (connected via a geographical proximity edge, diffusion coefficient 0.3). For large-scale power grid topology graphs, the number of neighboring nodes may be large, requiring efficient data structures and query algorithms to ensure rapid acquisition of neighbor information in iterative calculations at each time step.

[0224] After acquiring all neighboring nodes, it is necessary to calculate the difference between the potential energy value of each neighboring node at the current time step and the potential energy value of the target node at the current time step. This difference is called the potential energy gradient, reflecting the driving force for the propagation of potential energy between the two nodes. In actual implementation, for the target node i and its neighboring node j, their potential energy values ​​at the current time step t are read and denoted as follows: and Calculate the difference in potential energy values. .

[0225] The sign of the potential gradient has a clear physical meaning: when When the potential energy of neighboring node j is higher than that of node i, the potential energy will flow from node j to node i, and node i will receive the potential energy increment from node j; when When the potential energy of node i is higher than that of its neighbor node j, the potential energy will flow from node i to node j, node i will output potential energy to node j, and its own potential energy will decrease; when When the potential energy of the two nodes is equal, they are in equilibrium and no potential energy transfer occurs.

[0226] For example, at a certain time step, the potential energy value of node A is 50, the potential energy value of its neighbor node B is 80, the potential energy value of node C is 30, and the potential energy value of node D is 50. Then the calculated potential energy gradients are as follows:

[0227] ; ; ;

[0228] This means that node A will receive potential energy from node B, output potential energy to node C, and there will be no potential energy exchange with node D. The magnitude of the potential energy gradient determines the intensity of potential energy transfer; the larger the gradient, the faster the transfer rate.

[0229] After calculating the potential energy gradient, the difference in potential energy values ​​needs to be multiplied by the diffusion coefficient of the corresponding edge to obtain the potential energy transfer amount. This transfer amount represents the magnitude of potential energy transferred to the target node through that edge within one time step. In practical implementation, for target node i and its neighbor node j, the formula for calculating the potential energy transfer amount is: ;in, This represents the diffusion coefficient of the edge connecting node j and node i; This represents the potential energy gradient. The diffusion coefficient, as a modulator of propagation efficiency, determines the proportion of the potential energy gradient that can be converted into actual transmitted energy. The larger the diffusion coefficient, the greater the transmitted energy from the same potential energy gradient, and the faster the potential energy diffuses.

[0230] For example, if the potential energy gradient between node A and its neighbor node B is 30, and the diffusion coefficient of the connecting edge is 0.85, then the potential energy transfer is... The potential energy gradient between node A and its neighbor node C is -20, and the diffusion coefficient of the connecting edge is 0.75. Therefore, the potential energy transfer is... A negative value indicates that potential energy flows out of node A. The potential energy gradient between node A and its neighbor node D is 0, and the potential energy transfer is 0 regardless of the diffusion coefficient.

[0231] By introducing the diffusion coefficient, the model can distinguish the propagation efficiency of different propagation paths. Electrical topology edges, due to their higher diffusion coefficients, generate a larger transmission amount under the same potential energy gradient, reflecting the characteristic of rapid fault propagation along electrical lines. Geographically adjacent edges, due to their smaller diffusion coefficients, have a relatively smaller transmission amount, reflecting the slow propagation in geographic space.

[0232] After calculating the potential energy transfer between the target node and all its neighboring nodes, these transfer amounts need to be summarized, and the potential energy value of the target node in the next time step needs to be updated. In practice, for the target node i, all its neighboring nodes j are traversed, and the potential energy transfer amount from each neighboring node to node i is accumulated to obtain the total potential energy increment:

[0233] ;

[0234] in, Let represent the set of neighboring nodes of node i. The total potential energy increment reflects the net potential energy received by node i from its surroundings at the current time step; a positive value indicates that node i's potential energy will increase, and a negative value indicates that node i's potential energy will decrease. Then, the total potential energy increment is added to the potential energy value of node i at the current time step t. We obtain the potential energy value of node i at the next time step t+1:

[0235] ;

[0236] For example, if node A has a potential energy value of 50 at the current time step, receives 25.5 potential energy from neighbor node B, outputs -15 potential energy to neighbor node C, and has 0 potential energy transferred with neighbor node D, then the total potential energy increment is 25.5 + (-15) + 0 = 10.5. The potential energy value of node A at the next time step is 50 + 10.5 = 60.5.

[0237] It is important to note that for nodes defined as potential energy sources, their potential energy value remains constant during the iteration process, unaffected by the potential energy transfer from surrounding nodes. They consistently maintain their initially set high potential energy value and continuously output potential energy to their surroundings. For nodes defined as low potential wells, although their initial potential energy value is negative, they still update according to the above formula during the iteration process, gradually increasing their own potential energy value by absorbing the potential energy of surrounding nodes until they reach an equilibrium state.

[0238] This embodiment describes the calculation process of potential energy diffusion in detail, achieving an accurate numerical simulation of the propagation of fault potential energy in the power grid topology. All neighboring nodes of a node are obtained, and the diffusion coefficients of the connecting edges are acquired simultaneously, providing complete topological information and propagation parameters for calculating the potential energy transfer. The potential energy gradient is obtained by calculating the difference in potential energy values ​​between neighboring nodes and the target node, quantifying the driving force of potential energy propagation. The sign of the potential energy gradient clarifies the direction of potential energy flow, and the magnitude of the gradient determines the intensity of propagation, ensuring that the simulation process conforms to physical laws.

[0239] Multiplying the potential energy gradient by the diffusion coefficient yields the potential energy transfer, transforming the propagation driving force into the actual transfer amount. The diffusion coefficient, as a regulating factor, reflects the differences in propagation efficiency across different propagation paths, causing electrical topology edges and geographically proximate edges to produce different transfer effects under the same potential energy gradient, accurately reflecting the actual characteristics of fault propagation. Summing the potential energy transfer amounts of all neighboring nodes and updating the potential energy value of the target node completes the potential energy diffusion calculation for one time step. Iterative advancement realizes the dynamic evolution of the potential energy distribution. This calculation method exhibits good numerical stability and convergence, can handle complex power grid topologies, and is applicable to power grid systems of different scales. It provides a scientific basis for dynamic grid adjustment and emergency resource scheduling, significantly improving the accuracy and timeliness of fault response.

[0240] In one embodiment of this invention, the power grid drone inspection system further includes a ground base station connected to the drone cluster. The ground base station generates and executes collaborative inspection tasks for each drone cluster within a new grid based on the new grid boundary scheme, including the following steps:

[0241] S710. Analyze the new grid boundary scheme to obtain the boundary range of each new grid and the UAV resources within the grid, and determine the UAV endurance based on the UAV resources;

[0242] S720. Based on the real-time fault potential field distribution map, identify high-risk areas in each new grid where the potential energy value exceeds the preset potential energy threshold, and determine the potential energy gradient of the high-risk areas.

[0243] S730: For each new grid, based on the potential energy gradient of high-risk areas and the drone's endurance, a collaborative inspection path is planned for the drone cluster within the new grid.

[0244] The S740 encapsulates the collaborative inspection path into inspection task instructions and sends them to the corresponding drone cluster via ground base stations.

[0245] In practical implementation, the power grid drone inspection system includes a ground base station as the core communication hub. The ground base station maintains a real-time connection with the drone swarm via a wireless communication network, responsible for issuing task commands, monitoring flight status, and transmitting inspection data. Upon receiving a new grid boundary scheme, the system first needs to parse the scheme to extract key information. The new grid boundary scheme is stored in a structured data format, containing a unique identifier for each grid, a sequence of boundary coordinate points, a list of included inspection points, and a list of assigned drones.

[0246] The analysis process constructs the geographic polygon boundary of each new grid by reading the boundary coordinate point sequence, thus defining the spatial range covered by the grid. For example, the boundary of a new grid is defined by 8 coordinate points, forming an irregular polygon with a coverage area of ​​approximately 50 square kilometers. Simultaneously, it extracts drone resource information within the grid, including detailed parameters such as the drone's serial number, model, current location coordinates, remaining battery percentage, maximum flight speed, and the type of sensors it carries.

[0247] The range of each drone is determined based on these parameters, which are primarily determined by the remaining battery power and energy consumption characteristics. For electric multi-rotor drones, the maximum range can be obtained by dividing the remaining battery power by the energy consumption per unit distance. For example, if a drone has 80% remaining battery power and a full-charge range of 30 kilometers, its current range is 24 kilometers. Range calculations also need to include a safety margin, typically reserving 20% ​​of the battery power for return and emergency backup. Therefore, the actual usable range for inspection missions is 80% of the calculated value.

[0248] After determining the grid boundaries and UAV resources, it is necessary to identify high-risk areas within each new grid based on the real-time fault potential field distribution map. In practice, the real-time fault potential field distribution map is first spatially overlaid with the new grid boundary to extract all potential energy data falling within that grid boundary. Then, a preset potential energy threshold is set as the criterion for determining high-risk areas. This threshold is usually determined based on the statistical characteristics of the potential energy distribution, such as selecting the 75th or 80th quantile of the potential energy value as the threshold.

[0249] For example, if the potential energy value of the entire power grid ranges from 0 to 100, the potential energy threshold can be set to 60, indicating that areas with potential energy values ​​exceeding 60 are considered high-risk areas and require priority inspection. By traversing all potential energy data points within the grid, data points with potential energy values ​​greater than the threshold are selected and clustered geographically to form several consecutive high-risk areas. For example, three high-risk areas might be identified within a new grid, located in the north, middle, and south of the grid, with each area containing several inspection points.

[0250] After identifying high-risk areas, it is also necessary to determine the potential energy gradient of these areas. The potential energy gradient indicates the direction in which the potential energy increases most rapidly, which is also the direction in which the failure risk increases most rapidly. The potential energy gradient is calculated using a numerical differentiation method. For each location point within the high-risk area, the rate of change of potential energy in the east-west and north-south directions is calculated and then synthesized into a gradient vector. The direction of the gradient vector points to the direction in which the potential energy increases most rapidly, and the magnitude of the gradient vector represents the rate of potential energy increase. For example, if the potential energy gradient of a certain high-risk area points northeast and has a magnitude of 15, it means that moving a unit distance in the northeast direction increases the potential energy value by 15.

[0251] After identifying high-risk areas and determining the potential energy gradient, for each new grid, a collaborative inspection path is planned for the drone swarm within the new grid based on the potential energy gradient of the high-risk area and the drone's endurance. In actual implementation, the goal of path planning is to maximize the inspection coverage of high-risk areas while meeting the drone's endurance constraints, and to prioritize inspecting areas with the largest potential energy gradient.

[0252] First, the inspection points within the high-risk area are prioritized based on their potential energy gradients. Inspection points with larger potential energy gradients have higher priority and should be inspected first. For example, if a high-risk area contains 5 inspection points, the priority inspection order based on the potential energy gradient would be P3, P1, P5, P2, and P4. Then, a greedy or heuristic algorithm is used for path allocation. Inspection points are selected sequentially according to the priority order. For each inspection point, the flight distance from the current position to that point for all UAVs within the grid is calculated.

[0253] The flight distance is calculated considering terrain obstacles and no-fly zones, using a three-dimensional path planning algorithm to obtain the feasible flight distance. The flight distance is compared with the remaining range of the drones to filter candidate drones that meet the range requirements. The drone with the shortest flight distance is selected from the candidate drones, and the inspection point is assigned to that drone, updating its remaining range. This assignment process is repeated until all inspection points have been assigned or all drones have exhausted their remaining range.

[0254] For each drone, all inspection points assigned to it are sorted according to the principle of shortest flight distance to form a complete inspection path for that drone. Path planning also needs to consider the coordination between drones to avoid conflicts between multiple drones in the same area. Coordination can be achieved by setting time windows or spatial isolation. For example, if the inspection paths of drone A and drone B intersect in a certain area, drone A can be scheduled to complete the inspection of that area first, and drone B can enter 5 minutes later.

[0255] After completing the collaborative inspection path planning, the path information needs to be encapsulated into standardized inspection task instructions, which are then sent to the corresponding UAV cluster for execution via ground base stations. In actual implementation, the inspection task instructions adopt a structured data format, including detailed information such as task identifier, UAV number, task priority, waypoint coordinate sequence, flight altitude, flight speed, hovering time, sensor operating mode, and data acquisition parameters for each waypoint.

[0256] The waypoint coordinate sequence defines the complete flight trajectory of the UAV, with each waypoint containing three coordinate values: longitude, latitude, and altitude. Flight altitude is set according to the type of object being inspected and the terrain conditions. For power line inspection, the flight altitude is typically set to 20 to 50 meters from the conductor; for substation inspection, the flight altitude can be set to 50 to 100 meters. Flight speed is set according to the required inspection accuracy; slower speeds (e.g., 5 m / s) are used for detailed inspections, while faster speeds (e.g., 10 m / s) are used for routine inspections. Hovering time refers to the time the UAV spends at each waypoint collecting data, typically set to 5 to 30 seconds. Sensor operating modes include visible light imaging, infrared thermal imaging, and lidar scanning, selected according to the specific needs of the inspection task.

[0257] After the mission instructions are encapsulated, they are sent to the corresponding UAVs via the communication module of the ground base station. Communication between the ground base station and the UAV uses a 4G / 5G cellular network or a dedicated radio link to ensure real-time and reliable instruction transmission. Upon receiving the mission instructions, the UAV automatically parses the instructions, plans its takeoff time, performs pre-flight self-checks, and then autonomously flies according to the waypoint sequence in the instructions to complete the inspection mission. During flight, the UAV transmits flight status data and inspection data back to the ground base station in real time. The ground base station monitors the mission progress and can send correction instructions or emergency recall instructions when necessary.

[0258] This embodiment achieves a complete closed loop from grid adjustment to task execution through ground base stations, effectively improving the response speed and execution efficiency of emergency inspections. Analyzing the new grid boundary scheme and determining the UAV's endurance provides accurate resource constraints for path planning, ensuring that the planned inspection path is within the UAV's range and avoiding the risk of battery depletion during task execution. Based on the real-time fault potential field distribution map, high-risk areas are identified and potential gradients are calculated, enabling accurate identification of key inspection areas and prediction of risk evolution trends, allowing inspection resources to be focused on the areas requiring the most attention.

[0259] Based on potential energy gradient and endurance, a collaborative inspection path is planned. A greedy algorithm and heuristic optimization are employed to maximize inspection coverage in high-risk areas while meeting endurance constraints. Priority is given to areas with the largest potential energy gradient, ensuring timely fault detection. Coordination between drones is considered, and time windows or spatial isolation are set to avoid conflicts between multiple drones in the same area, improving the safety and orderliness of the inspection mission. The collaborative inspection path is encapsulated into standardized task instructions and issued via ground base stations, enabling rapid deployment and automated execution, reducing the need for manual intervention. Drones transmit flight status and inspection data in real time, and ground base stations continuously monitor mission progress, sending correction commands when necessary. This enhances the system's flexibility and controllability, significantly shortens the time interval from fault occurrence to inspection implementation, and improves the overall effectiveness of power grid fault response.

[0260] In one embodiment of this invention, based on the potential energy gradient of the high-risk area and the drone's endurance, a collaborative inspection path is planned for the drone swarm within the new grid, including the following steps:

[0261] S810. Based on the potential energy gradient in high-risk areas, determine the direction of the fastest increase in potential energy.

[0262] S820: Determine the priority inspection sequence based on the direction of the fastest increase in potential energy.

[0263] S830: Obtain the current location, remaining battery power, and maximum range of all drones within the new grid.

[0264] S840. Select inspection points in order of priority inspection and calculate the flight distance of each drone from its current position to the inspection point.

[0265] S850: Repeat the path planning steps until all inspection points are assigned or all drones run out of battery life, generating a complete inspection path for each drone in the drone cluster.

[0266] The path planning steps include:

[0267] For the current inspection point, the flight distance is compared with the maximum range of the drone to select candidate drones whose range meets the requirements;

[0268] Select the drone with the shortest flight distance from the candidate drones, assign the inspection point to that drone, and update the drone's remaining range.

[0269] In practical implementation, determining the direction of the fastest potential energy increase based on the potential energy gradient in the high-risk area is a crucial starting point for path planning. The potential energy gradient is a vector field, and there is a corresponding gradient vector at each spatial location in the high-risk area. The direction of this vector points to the direction of the fastest potential energy increase, and the magnitude of the vector represents the rate of potential energy increase.

[0270] For each inspection point within a high-risk area, the potential energy gradient vector at that point is first obtained. The gradient vector is typically represented in two dimensions, containing east-west and north-south components. For example, the potential energy gradient vector of inspection point P1 is (12, 9), indicating that the rate of change of potential energy along the east direction is 12, and the rate of change of potential energy along the north direction is 9. By calculating the direction angle of the gradient vector, the direction of the fastest increase in potential energy can be determined. The direction angle is calculated using the arctangent function for the gradient vector. The direction angle is This angle represents the deflection angle relative to due east.

[0271] For example, for the gradient vector (12, 9), the direction angle is approximately 36.87 degrees, indicating that the direction of the fastest increase in potential energy is 36.87 degrees east of north, i.e., northeast. For all inspection points within a high-risk area, the direction of the fastest increase in potential energy is calculated and statistically analyzed. If the potential energy gradient directions of most inspection points tend to be consistent, it indicates that the fault risk is spreading rapidly along a dominant direction, forming a clear fault propagation zone. For example, if a high-risk area contains 8 inspection points, and the potential energy gradient directions of 6 of these points all point northeast, it indicates that the fault is spreading rapidly along the northeast direction.

[0272] By identifying this dominant propagation direction, the evolution trend of faults can be predicted, providing forward-looking guidance for inspection path planning. Furthermore, it is necessary to calculate the magnitude of the potential energy gradient; a larger magnitude indicates a faster increase in potential energy, a faster rise in fault risk in that area, and a greater need for an urgent inspection response.

[0273] After determining the direction of the fastest increase in potential energy, a priority inspection sequence is established based on this direction. This ensures that the UAV can conduct proactive inspections along the direction of fault propagation, promptly detecting and preventing the spread of faults. In practice, the priority inspection sequence is determined using a comprehensive scoring mechanism, taking into account both the magnitude and directional consistency of the potential energy gradient.

[0274] First, for each inspection point within a high-risk area, calculate its potential energy gradient magnitude. A larger magnitude indicates a faster increase in fault risk at that point, thus a higher priority. Second, calculate the angle between the potential energy gradient direction and the dominant propagation direction at each inspection point. A smaller angle indicates that the point is closer to the forefront of fault propagation, thus a higher priority. For example, if the dominant propagation direction is northeast (45 degrees), the gradient direction of inspection point P1 is northeast-east (50 degrees), with an angle of 5 degrees; the gradient direction of inspection point P2 is due north (90 degrees), with an angle of 45 degrees. Clearly, inspection point P1 is closer to the dominant propagation direction and should be inspected first.

[0275] The comprehensive scoring formula can be designed as follows: the priority score equals the potential energy gradient magnitude multiplied by the weight coefficient w1, minus the direction angle multiplied by the weight coefficient w2. Here, w1 and w2 are set according to actual needs; typically, w1 is set to 0.7 and w2 to 0.3, reflecting the dominant role of gradient magnitude. After calculating the priority scores of all inspection points, they are sorted from highest to lowest score to obtain the priority inspection order.

[0276] After determining the priority inspection order, it is necessary to obtain the real-time status information of all drones within the new grid to provide resource constraints for path planning. In actual implementation, key parameters such as the current location, remaining battery power, and maximum range of each drone are queried in real time through the communication link between the ground base station and the drone cluster. The current location is represented in geographic coordinates, including longitude, latitude, and altitude. For example, a drone is currently located at 118.52 degrees east longitude, 32.35 degrees north latitude, and an altitude of 150 meters.

[0277] Remaining battery power is expressed as a percentage, obtained by reading data from the drone's battery management system; for example, a drone may have 75% remaining battery power. Maximum range refers to the maximum distance a drone can fly on a full charge. This parameter is determined by the drone's model and configuration and is stored in the drone's technical parameter database. For example, a quadcopter drone may have a full-charge range of 30 kilometers, while a fixed-wing drone may have a full-charge range of up to 80 kilometers.

[0278] Based on the remaining battery power and maximum range, the current available range for each drone is calculated as follows: available range equals maximum range multiplied by the remaining battery percentage, then multiplied by a safety factor. The safety factor is typically set to 0.8, reserving 20% ​​battery power for return and emergency backup, ensuring the drone can safely return to its takeoff point or the nearest landing point. For example, if a drone has a maximum range of 30 kilometers and 75% remaining battery power, its available range is 30 × 0.75 × 0.8 = 18 kilometers.

[0279] For all drones within the grid, these parameters are acquired and calculated one by one to form a drone resource status table. For example, if a new grid is equipped with 5 drones, the resource status table is recorded as follows: Drone U1 (location A, remaining battery 80%, available range 19.2 km), Drone U2 (location B, remaining battery 65%, available range 15.6 km), Drone U3 (location C, remaining battery 90%, available range 21.6 km), Drone U4 (location D, remaining battery 55%, available range 13.2 km), and Drone U5 (location E, remaining battery 70%, available range 16.8 km).

[0280] After acquiring the drone resource status, inspection points are selected sequentially according to priority, and path planning is performed to allocate drones. In practice, the first inspection point, such as inspection point P3, is selected from the priority inspection list. For this inspection point, the flight distance from the current position to that point for each drone within the grid needs to be calculated. The flight distance calculation is not a simple straight-line distance, but rather takes into account terrain obstacles, no-fly zone restrictions, and the actual feasible flight path length for safe flight altitude.

[0281] Using a three-dimensional path planning algorithm, such as the A* algorithm or the fast expanding random tree algorithm, the shortest feasible path from the UAV's current position to the target inspection point is calculated based on the digital elevation model and airspace constraint data. For example, the flight distance of UAV U1 from position A to inspection point P3 is 8.5 km, that of UAV U2 from position B to P3 is 12.3 km, that of UAV U3 from position C to P3 is 6.2 km, that of UAV U4 from position D to P3 is 15.8 km, and that of UAV U5 from position E to P3 is 9.7 km.

[0282] The next step is path planning, which involves two key operations. First, for the current inspection point P3, the flight distance of each drone to this point is compared with its available range to filter candidate drones that meet the range requirements. It's important to note that the flight distance should include the round-trip distance, i.e., the total distance from the current location to the inspection point and back, typically estimated as twice the flight distance. For example, drone U1's flight distance to P3 is 8.5 kilometers, and its round-trip distance is 17 kilometers, while U1's available range is 19.2 kilometers, meeting the requirement and being included as a candidate drone. Drone U2's round-trip distance is 24.6 kilometers, exceeding its available range by 15.6 kilometers, not meeting the requirement and being excluded. After screening, the candidate drones are U1, U3, and U5.

[0283] Next, select the drone with the shortest flight distance from the candidate drones and assign the inspection point to that drone. In this example, drone U3 has the shortest flight distance from P3, which is 6.2 kilometers, so inspection point P3 is assigned to drone U3. After the assignment, the status information of drone U3 needs to be updated, including updating the current location to P3, reducing the remaining range by 12.4 kilometers (round trip distance), and updating the available range to 9.2 kilometers.

[0284] Repeat the path planning steps described above, processing the next inspection point in the priority inspection order in turn. For example, when processing inspection point P1, calculate the flight distance from all drones to P1, filter candidate drones, select the drone with the shortest flight distance for assignment, and update the status of that drone. The iterative process continues until the termination condition is met: all inspection points have been assigned, or all drones have run out of battery life and cannot continue to perform the task.

[0285] During the iteration process, some drones may be assigned multiple inspection points, forming an inspection path containing multiple waypoints; some drones may not be assigned any inspection points due to remote locations or insufficient endurance. Ultimately, a complete inspection path is generated for each drone in the grid, represented as a sequence of waypoints. For example, the inspection path for drone U3 is: takeoff point C, P3, P1, P5, return point C.

[0286] This embodiment utilizes an intelligent path planning method based on potential energy gradients to achieve proactive inspection and priority coverage of high-risk areas, significantly improving the timeliness of fault detection and the utilization efficiency of inspection resources. By determining the direction of the fastest increase in potential energy based on the potential energy gradient, the dominant direction of fault propagation is identified, providing a scientific basis for predicting fault evolution trends and enabling proactive deployment of inspection paths along the fault propagation direction.

[0287] Priority inspection order is determined based on the magnitude and direction consistency of the potential energy gradient. A comprehensive scoring mechanism is adopted, taking into account both the urgency of the risk and the location of the propagation front, ensuring that the highest-risk and most urgent areas receive priority inspection and maximizing the effectiveness of limited inspection resources. Real-time status information of the UAV is acquired and available range is calculated, providing accurate resource constraints for path planning. A safety margin is reserved to ensure the UAV can return safely, improving the reliability of mission execution.

[0288] Inspection points are assigned sequentially according to priority. A greedy strategy is used to select the drone with the shortest flight distance, optimizing flight efficiency while meeting endurance constraints and reducing drone energy consumption and task execution time. The path planning steps are repeatedly executed, and drone status is dynamically updated, enabling collaborative task allocation among multiple drones and avoiding resource idleness and task omissions. The generated complete inspection path is represented as a waypoint sequence, containing detailed flight trajectory information, facilitating autonomous drone execution and ground base station monitoring. This allows for the detection and handling of potential hazards before a fault fully spreads, effectively preventing cascading fault propagation and significantly improving the power grid's safe operation and emergency response capabilities.

[0289] See Figure 3 The present invention also provides a multi-level gridded deployment system based on hierarchical clustering, the system being used to implement the aforementioned multi-level gridded deployment method based on hierarchical clustering, the system comprising:

[0290] The initial inspection grid acquisition module is used to acquire digital elevation model data of the inspection area in the power grid inspection points; for any pair of inspection points, it calculates the three-dimensional geographic surface flight distance and calculates the communication signal attenuation prediction value based on the digital elevation model data and the preset wireless channel propagation model; based on the three-dimensional geographic surface flight distance and the communication signal attenuation prediction value, it constructs a multi-dimensional flight cost matrix; and aggregates all inspection points using the multi-dimensional flight cost matrix as input to obtain the initial inspection grid.

[0291] The real-time fault potential field distribution map acquisition module is used to acquire equipment fault alarm information in real time, and based on the equipment fault alarm information, determine the power grid inspection point where the fault occurred, the risk equipment, and the normal equipment; define the power grid inspection point where the fault occurred as a potential energy source and set an initial potential energy value, define the risk equipment as a low potential well and set a potential energy value, and set the potential energy of the normal equipment to 0; acquire the electrical topology and geographical proximity of the power grid inspection point, and generate a real-time fault potential field distribution map based on the electrical topology and geographical proximity to simulate the process of potential energy spreading from the potential energy source to the surrounding area;

[0292] The new inspection grid acquisition module is used to calculate the total fault entropy value of the current power grid system based on the real-time fault potential field distribution map, evaluate the predefined grid operation set based on the total fault entropy value, select the operation that maximizes the entropy deceleration rate as the optimal decision, and output the adjusted new grid boundary scheme; according to the new grid boundary scheme, it generates collaborative inspection tasks for the UAV cluster in each new grid and issues them for execution.

[0293] The present invention also provides a power grid unmanned aerial vehicle (UAV) inspection system, comprising:

[0294] Central control host;

[0295] A drone swarm, connected to a central control unit;

[0296] The ground base station is connected to both the central control host and the drone swarm.

[0297] See Figure 4 The present invention also provides a multi-level gridded deployment device based on hierarchical clustering, including a memory and a processor;

[0298] The memory is used to store computer program code and transmit the computer program code to the processor;

[0299] The processor is configured to execute, according to instructions in the computer program code, a multi-level gridded deployment method based on hierarchical clustering as described above.

[0300] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a multi-level gridded deployment method based on hierarchical clustering as described above.

[0301] Generally, the computer instructions for implementing the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself, which is temporarily propagating.

[0302] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EKROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0303] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks such as TensorFlow and PyTorch can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer or to an external computer (e.g., via the Internet using an Internet service provider) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0304] The aforementioned equipment and non-transitory computer-readable storage media can be found in the detailed description of a multi-level gridded deployment method based on hierarchical clustering and its beneficial effects, which will not be repeated here.

[0305] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A multi-level gridded deployment method based on hierarchical clustering, characterized in that, Applied to a power grid drone inspection system, which includes a drone swarm, the method includes: Obtain digital elevation model data of the inspection area in the power grid inspection points; For any pair of inspection points, calculate the flight distance on the three-dimensional geographic surface, and calculate the predicted value of communication signal attenuation based on digital elevation model data and a preset wireless channel propagation model. A multidimensional flight cost matrix is ​​constructed based on the predicted flight distance and communication signal attenuation values ​​of three-dimensional geographic surface. The multidimensional flight cost matrix is ​​used as input to aggregate all inspection points to obtain the initial inspection grid. Real-time acquisition of equipment fault alarm information, and based on the equipment fault alarm information, identification of power grid inspection points where faults have occurred, risky equipment, and normal equipment; Define the faulty power grid inspection point as a potential energy source and set an initial potential energy value; define the risky equipment as a low potential sink and set a potential energy value; set the potential energy of normal equipment to 0. The electrical topology and geographical proximity of power grid inspection points are obtained, and a real-time fault potential field distribution map is generated based on the electrical topology and geographical proximity to simulate the process of potential energy spreading from the potential energy source to the surrounding area. Based on the real-time fault potential field distribution map, the total fault entropy value of the current power grid system is calculated, and the predefined set of grid operations is evaluated based on the total fault entropy value. The operation that maximizes the entropy deceleration rate is selected as the optimal decision, and the adjusted new grid boundary scheme is output. Based on the new grid boundary scheme, collaborative inspection tasks are generated for each drone cluster within the new grid and then issued for execution.

2. The multi-level gridded deployment method based on hierarchical clustering partitioning according to claim 1, characterized in that, The multidimensional flight cost matrix is ​​constructed based on the predicted flight distance and communication signal attenuation values ​​from the three-dimensional geographic surface, including: For any pair of inspection points, the first weighted value is obtained by multiplying the flight distance of the three-dimensional geographic surface by the preset flight distance weighting coefficient, and the second weighted value is obtained by multiplying the predicted value of communication signal attenuation by the preset communication attenuation weighting coefficient. The sum of the first weighted value and the second weighted value yields the comprehensive flight cost of the inspection point pair; The comprehensive flight costs of all inspection point pairs are organized into a matrix form to obtain a multidimensional flight cost matrix.

3. The multi-level gridded deployment method based on hierarchical clustering partitioning according to claim 1, characterized in that, The process of aggregating all inspection points using a multidimensional flight cost matrix as input to obtain an initial inspection grid includes: Initialize the clustering state, treating each inspection point as an independent cluster; Repeat the aggregation step until the number of clusters reaches the preset grid number threshold; Each resulting cluster is defined as the initial inspection grid. The polymerization step includes: Calculate the distance between any two clusters based on the multidimensional flight cost matrix; The two clusters with the smallest distance are selected and merged to form a new cluster; Update the multidimensional flight cost matrix, recalculate the inter-cluster distances that include the new clusters, and determine the number of current clusters based on the inter-cluster distances.

4. The multi-level gridded deployment method based on hierarchical clustering partitioning according to claim 1, characterized in that, The process of identifying faulty power grid inspection points, at-risk equipment, and normal equipment based on equipment fault alarm information includes: Analyze the equipment fault alarm information to obtain the location information and fault type of the faulty equipment; The fault level is queried from the preset fault level mapping table according to the fault type, and the power grid inspection point where the fault occurred is determined based on the fault level. Obtain the operating years and historical fault frequency of each device at the power grid inspection point where a fault occurred, and calculate the risk score of each device; Devices with risk scores exceeding a preset risk threshold are identified as risky devices, while devices with risk scores below the preset risk threshold are identified as normal devices.

5. The multi-level gridded deployment method based on hierarchical clustering partitioning according to claim 1, characterized in that, The generation of a real-time fault potential field distribution map based on electrical topology and geographical proximity includes: Construct a power grid topology graph, where nodes represent power grid inspection points and edges represent electrical topology or geographical proximity relationships; Assign a diffusion coefficient to each edge of the power grid topology graph, where the diffusion coefficient corresponding to electrical topology is greater than the diffusion coefficient corresponding to geographical proximity. Based on the initial potential energy value and diffusion coefficient, calculate the potential energy value of each node in the power grid topology at each time step; The iteration stops when the number of iterations reaches the preset iteration threshold or the potential energy distribution converges, and the final potential energy distribution is obtained. The final potential energy distribution is mapped to a preset power grid geographic space to generate a real-time fault potential field distribution map.

6. The multi-level gridded deployment method based on hierarchical clustering partitioning according to claim 5, characterized in that, The calculation of the potential energy value of each node in the power grid topology at each time step, based on the initial potential energy value and the diffusion coefficient, includes: For any node, obtain all neighboring nodes that are adjacent to that node; Calculate the difference between the potential energy value of the neighboring node at the current time step and the potential energy value of this node at the current time step; Multiply the difference in potential energy values ​​by the diffusion coefficient of the corresponding edge to obtain the potential energy transfer amount; Sum the potential energy transferred from all neighboring nodes to this node, and add the potential energy value of this node at the current time step to obtain the potential energy value of this node at the next time step.

7. The multi-level gridded deployment method based on hierarchical clustering partitioning according to claim 1, characterized in that, The power grid drone inspection system also includes a ground base station, which is connected to the drone swarm. Based on the new grid boundary scheme, the ground base station generates and executes collaborative inspection tasks for each drone swarm within the new grid, including: The new grid boundary scheme is analyzed to obtain the boundary range of each new grid and the UAV resources within the grid, and the UAV endurance is determined based on the UAV resources; Based on the real-time fault potential field distribution map, high-risk areas with potential energy values ​​exceeding the preset potential energy threshold are identified in each new grid, and the potential energy gradient of the high-risk areas is determined. For each new grid, based on the potential energy gradient of high-risk areas and the drone's endurance, a collaborative inspection path is planned for the drone swarm within the new grid; The collaborative inspection path is encapsulated as an inspection task instruction and sent to the corresponding drone cluster via ground base stations.

8. A multi-level gridded deployment method based on hierarchical clustering partitioning according to claim 7, characterized in that, The method of planning collaborative inspection paths for drone swarms within the new grid, based on the potential energy gradient of high-risk areas and the drone's endurance, includes: Based on the potential energy gradient in high-risk areas, determine the direction of the fastest increase in potential energy. Determine the priority inspection order based on the direction of the fastest increase in potential energy. Obtain the current location, remaining battery power, and maximum range of all drones within the new grid; Select inspection points in order of priority and calculate the flight distance of each drone from its current position to the inspection point. Repeat the path planning steps until all inspection points have been assigned or all drones have run out of battery life, generating a complete inspection path for each drone in the drone swarm. The path planning steps include: For the current inspection point, the flight distance is compared with the maximum range of the drone to select candidate drones whose range meets the requirements; Select the drone with the shortest flight distance from the candidate drones, assign the inspection point to that drone, and update the drone's remaining range.

9. A multi-level gridded deployment system based on hierarchical clustering, characterized in that, The system is used to implement the method according to any one of claims 1 to 8, the system comprising: The initial inspection grid acquisition module is used to acquire digital elevation model data of the inspection area in the power grid inspection points; for any pair of inspection points, it calculates the three-dimensional geographic surface flight distance and calculates the communication signal attenuation prediction value based on the digital elevation model data and the preset wireless channel propagation model; based on the three-dimensional geographic surface flight distance and the communication signal attenuation prediction value, it constructs a multi-dimensional flight cost matrix; and aggregates all inspection points using the multi-dimensional flight cost matrix as input to obtain the initial inspection grid. The real-time fault potential field distribution map acquisition module is used to acquire equipment fault alarm information in real time, and based on the equipment fault alarm information, determine the power grid inspection point where the fault occurred, the risk equipment, and the normal equipment; define the power grid inspection point where the fault occurred as a potential energy source and set an initial potential energy value, define the risk equipment as a low potential well and set a potential energy value, and set the potential energy of the normal equipment to 0; acquire the electrical topology and geographical proximity of the power grid inspection point, and generate a real-time fault potential field distribution map based on the electrical topology and geographical proximity to simulate the process of potential energy spreading from the potential energy source to the surrounding area; The new inspection grid acquisition module is used to calculate the total fault entropy value of the current power grid system based on the real-time fault potential field distribution map, evaluate the predefined grid operation set based on the total fault entropy value, select the operation that maximizes the entropy deceleration rate as the optimal decision, and output the adjusted new grid boundary scheme; according to the new grid boundary scheme, it generates collaborative inspection tasks for the UAV cluster in each new grid and issues them for execution.

10. A multi-level gridded deployment device based on hierarchical clustering, characterized in that, Including memory and processor; The memory is used to store computer program code and transmit the computer program code to the processor; The processor is configured to execute the method as described in any one of claims 1 to 8 according to instructions in the computer program code.

Images